Embed Size (px)
Citation preview
Online Allocation Algorithms with Applications in
Computational Advertising
by
Morteza Zadimoghaddam
Submitted to the Department of Electrical Engineering and ComputerScience
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy in Computer Science
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
February 2014
c© Massachusetts Institute of Technology 2014. All rights reserved.
Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Department of Electrical Engineering and Computer Science
January 24, 2014
Certified by. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Erik D. Demaine
ProfessorThesis Supervisor
Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Leslie A. Kolodziejski
Chairman, Department Committee on Graduate Students
To Faezeh
Online Allocation Algorithms with Applications in
Computational Advertising
by
Morteza Zadimoghaddam
Submitted to the Department of Electrical Engineering and Computer Scienceon January 24, 2014, in partial fulfillment of the
requirements for the degree ofDoctor of Philosophy in Computer Science
Abstract
Over the last few decades, a wide variety of allocation markets emerged from theInternet and introduced interesting algorithmic challenges, e.g., ad auctions, onlinedating markets, matching skilled workers to jobs, etc. I focus on the use of allocationalgorithms in computational advertising as it is the quintessential application of myresearch. I will also touch on the classic secretary problem with submodular utilityfunctions, and show that how it is related to advertiser’s optimization problem incomputational advertising applications. In all these practical situations, we shouldfocus on solving the allocation problems in an online setting since the input is beingrevealed during the course of the algorithm, and at the same time we should makeirrevocable decisions. We can formalize these types of computational advertisingproblems as follows. We are given a set of online items, arriving one by one, and aset of advertisers where each advertiser specifies how much she wants to pay for eachof the online items. The goal is to allocate online items to advertisers to maximizesome objective function like the total revenue, or the total quality of the allocation.There are two main classes of extensively studied problems in this context: budgetedallocation (a.k.a. the adwords problem) and display ad problems. Each advertiser isconstrained by an overall budget limit, the maximum total amount she can pay inthe first class, and by some positive integer capacity, the maximum number of onlineitems we can assign to her in the second class.
Thesis Supervisor: Erik D. DemaineTitle: Professor
3
Acknowledgments
As I am approaching the end of my doctorate studies at MIT, I vividly remember the
people who had a great influence on my life. I am indebted to them for their help
and support throughout this journey. At first I would like to thank my mother and
father (Fatemeh and Abbas); it is because of their never ending support that I have
had the chance to progress in life. Their dedication to my education provided the
foundation for my studies.
I started studying extracurricular Mathematics and Computer Science books in
high school with the hope of succeeding in the Iranian Olympiad in Informatics. For
this, I am grateful to Professor Mohammad Ghodsi; for the past many years he has
had a critical role in organizing Computer Science related competitions and training
camps all around Iran. With no doubt, without his support and guidance I would
not have had access to an amazing educational atmosphere.
Probably the main reason I got admitted to MIT was the excellent guidance of
Professor MohammadTaghi Hajiaghayi on how to conduct research in Theoretical
Computer Science. From my undergraduate years up to now, Prof. Hajiaghayi has
been a great colleague to work with. I would also like to thank Dr. Vahab Mirrokni
who has been a great mentor for me in my graduate studies and during my internship
at Google research. I am grateful for his support, advice, and friendship.
I would like to express my deepest gratitude to my advisor, Professor Erik Demaine
for providing me with an excellent environment for conducting research. Professor
Demaine has always amazed me by his intuitive way of thinking and his great teaching
skills. He has been a great support for me and a nice person to talk to.
I would like to thank my committee members, Professors Piotr Indyk and Costis
Daskalakis who were more than generous with their expertise and precious time.
I would also like to thank all my friends during my studies. In particular, I want
to thank Mohammad Norouzi, Mohammad Rashidian, Hamid Mahini, Amin Karbasi,
Hossein Fariborzi, Dan Alistarh, and Martin Demaine.
Finally, and most importantly, I want to thank my wife, Faezeh, who has been
4
very supportive during all my study years. Without your encouragements, I could
not come this far. You inspire me by the way you look at life. Thank you for making
all these years so wonderful.
5
Contents
1 Introduction 9
1.1 Budgeted Allocation Problem . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Display Ad Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Submodular Secretary Problem . . . . . . . . . . . . . . . . . . . . . 14
2 Simultaneous Algorithms for Budgeted Allocation Problem 16
2.1 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Main Ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.1 Lower bounding the Increase in the Potential Function . . . . 23
2.2.2 Description of the Factor-Revealing Mathematical Program . . 26
2.3 The Competitive Ratio of Weighted-Balance . . . . . . . . . . . . . . 29
2.4 The Competitive Ratio of Balance . . . . . . . . . . . . . . . . . . . 32
2.5 Hardness Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.6 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3 Bicriteria Online Matching: Maximizing Weight and Cardinality 41
3.1 Results and Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Hardness Instances . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.1 Better Upper Bounds via Factor-Revealing Linear Programs . 48
3.2.2 Hardness Results for Large Values of Weight Approximation
Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3 Algorithm for Large Capacities . . . . . . . . . . . . . . . . . . . . . 56
3.4 Algorithm for Small Capacities . . . . . . . . . . . . . . . . . . . . . 61
6
3.4.1 Lower Bounding the Weight Approximation Ratio . . . . . . . 62
3.4.2 Factor Revealing Linear Program for CardinalityAlg . . . . . . 63
4 Submodular Secretary Problem and its Extensions 66
4.1 Our Results and Techniques . . . . . . . . . . . . . . . . . . . . . . . 71
4.2 The Submodular Secretary Problem . . . . . . . . . . . . . . . . . . . 73
4.2.1 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 The Submodular Matroid Secretary Problem . . . . . . . . . . . . . . 82
4.4 Knapsack Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5 The Subadditive Secretary Problem . . . . . . . . . . . . . . . . . . . 89
4.5.1 Hardness Result . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.5.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.6 Further Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.6.1 The Secretary Problem with the “Maximum” Function . . . . 95
7
List of Figures
3-1 New curves for upper and lower bounds. . . . . . . . . . . . . . . . . 45
8
Chapter 1
Introduction
Over the last few decades, a wide variety of allocation markets have emerged. Some of
the most important examples of these markets are a) online dating markets that span
around 75% of single people in United States, b) matching skilled workers with jobs
such as National Resident Matching Program (NRMP) (or the Match) that matches
medical school students with residency programs, c) Ad Auction multi-billion dollars
markets which I elaborate more on it later as the quintessential application of my
research. The emergence of these markets introduced many interesting optimization
and algorithmic challenges. We can formalize these problems with a set of resources
present in advance that should be consumed with a set of online nodes arriving one
at a time during the algorithm. The goal is to design an algorithm that allocates
these resources to online nodes when they arrive without over exhausting resources.
We note that the allocation decisions are irrevocable. There are several objective
functions considered in these problems including a) the revenue achieved by these
allocations, and b) the quality of the allocation which is the total satisfaction of
online nodes based on the resources they have received. We have studied designing
online allocation algorithms that incorporate these multiple objectives simultaneously.
Another important aspect of these problems that contributes to their online nature
is the arrival order of online nodes. The performance of an online algorithm heavily
depends on which nodes are going to arrive and their arrival order. In the worst
case approach, the algorithm’s performance is measured assuming the arrival order
9
is determined by an adversary, and therefore the algorithm’s worst performance is
computed as its indicative performance. However in stochastic settings, the arrival
order is assumed to be a random permutation of online items, and the expected
value of algorithm’s performance is considered as its indicative performance. How to
combine these two settings, and coming up with a more realistic analysis framework
is the other main topic of my research.
I would like to highlight the use of allocation algorithms in computational advertis-
ing which is the quintessential application of my research e.g., the multi-billion dollar
ad auction markets of Google or Bing search engines. We can formalize these types
of computational advertising problems as follows. We are given a set of online items
that could represent keywords searched in a search engine during a day, pageviews
of a news website, or users of any website that shows ads. These online items are
arriving one by one, and the algorithm should serve these items by assigning them
some advertisements. On the other hand, we have a set of advertisers where each
advertiser specifies how much she wants to pay for each of the online items when they
arrive. There are multiple important objectives in this context including maximizing
the total revenue, the number of clicks on ads, or the number of served online items.
In these advertising applications, advertisers are in fact the resources, and there are
different types of constraints on these resources depending on the application. There
are two main classes of extensively studied problems in this context: budgeted allo-
cation (a.k.a. the adwords problem) and display ad problems. Each advertiser is
constrained by an overall budget limit, the maximum total amount she can pay in
the first class, and by some positive integer capacity, the maximum number of online
items we can assign to her in the second class.
The main issues of combining different objectives, and considering different arrival
orders are discussed above as part of the optimization problem the publisher is facing.
The publisher (Google, Bing, etc) is the search engine that decides which ads to show
for each online node that arrives and has to maintain some revenue and keep up the
quality (relevance) of these ads to maintain the online users in long term. At the end
of this thesis, we look at these settings from an advertisers perspective which has a
10
limited budget and tries to win (capture) the most relevant online nodes to publicize
her own business. We will see later how this is related to the classic secretary problem,
and how we should extend the classic secretary problem to capture complexities of
the advertiser’s problem in online ad auctions.
1.1 Budgeted Allocation Problem
The performance of online algorithms for the adwords (Budgeted Allocation) prob-
lem heavily depends on the order in which the online items will appear. Mehta et
al.[65] provided a 1 − 1/e competitive algorithm which works for every input order
(adversarial or worst case order). They presented a novel adaptive scoring function
on the advertisers (based on their remaining budgets) and allocated each online item
to an advertiser for whom the product of her bid for the online item and her score
is maximized. In another independent direction, researchers assumed that the online
items appear according to a random permutation. Devanur and Hayes [23] proposed
a Primal-Dual 1−ε competitive algorithm in this case which can be adapted to other
similar stochastic settings as well. The adversarial case is too pessimistic to model
reality. On the other hand, the random arrival (and other stochastic) setting is useful
only if the incoming traffic patterns of online items (e.g., page-views) can be predicted
with a reasonably good precision. In other words, such algorithms may rely heavily on
a precise forecast of the online traffic patterns, and may not react quickly to sudden
changes in these patterns. In fact, the slow reaction to such spikes in traffic patterns
imposes a serious limitation on the real-world use of stochastic algorithms in practical
applications. This is a common issue in applying stochastic optimization techniques
to the online resource allocation problems, e.g., see [79]. In an effort to tackle this ad
allocation problem, we [67] study algorithms that achieve good performance ratios
in both adversarial and stochastic frameworks simultaneously, and are robust against
different traffic patterns.
We present algorithms that achieve the best competitive ratios of both settings
simultaneously, i.e. 1− ε in the random arrival model and 1− 1/e in the adversarial
11
case for unweighted graphs. However, for weighted graphs we prove that this is not
possible; we show when the competitive ratio of an online algorithm tends to 1 in
the random arrival model, its competitive ratio tends to 0 in the adversarial model.
Formally, we prove that an algorithm with competitive ratio 1 − ε in the random
order arrival setting cannot have a competitive ratio more than 4√ε in the adversarial
setting. We also prove that no online algorithm that achieves an approximation factor
of 1 − 1/e for the adversarial inputs can achieve an average approximation factor
better than 97.6% for random arrival inputs. In light of this hardness result, we
design algorithms with improved approximation ratios in the random arrival model
while preserving the competitive ratio of 1 − 1/e in the worst case. To this end,
we show the algorithm proposed by [65] achieves a competitive ratio of 0.76 for the
random arrival model, while having a 1 − 1/e ≈ 0.63 competitive ratio in the worst
case.
Main Techniques: To achieve the hardness result, we exploit the fact that the
algorithm has no means of distinguishing between adversarial and stochastic inputs.
It is then sufficient to construct instances in which the behavior of an optimum online
algorithm in the stochastic setting differs drastically from an algorithm that works well
in the adversarial setting. For the positive results, we propose a general three stage
process which is useful in analyzing the performance of any greedy algorithm: (i) We
first define an appropriate potential function as the sum of the indefnite integrals of
the advertisers’ scores, and interpret the online algorithm as a greedy approach acting
to improve the potential function by optimizing the corresponding scoring functions.
(ii) We then track the changes in the potential function by formulating a mathematical
program. (iii) Finally, by discretization on two spectrums of time and budgets and
applying the mean value theorem of calculus, we translate the mathematical program
into a constant-size LP and solve it numerically. We prove that the solution of this
LP lower bounds the competitive ratio of the algorithm.
12
1.2 Display Ad Problem
In contrast to the budgeted allocation problem, in the display advertising problem,
advertisers are not constrained by budget caps. Instead, each advertiser has some
integer capacity which is the maximum number of online items we can assign to her.
This problem has been modeled as maximizing the weight of an online matching
instance [35, 34, 25, 15, 24]. While weight is indeed important, this model ignores
the fact that cardinality of the matching is also crucial in the display ad application.
This fact illustrates that in many real applications of online allocation, one needs to
optimize multiple objective functions, though most of the previous work in this area
deals with only a single objective function. On the other hand, there is a large body of
work exploring offline multi-objective optimization in the approximation algorithms
literature. In this part of thesis, we focus on simultaneously maximizing online two
objectives which have been studied extensively in matching problems: cardinality and
weight.
In online display advertising, advertisers typically purchase bundles of millions
of display ad impressions from web publishers. Display ad serving systems that
assign ads to pages on behalf of publishers must satisfy the contracts with advertisers,
respecting targeting criteria and delivery goals. Modulo this, publishers try to allocate
ads intelligently to maximize overall quality (measured, for example, by clicks), and
therefore a desirable property of an ad serving system is to maximize this quality
while satisfying the contracts to deliver to each advertiser its purchased bundle of
impressions. This motivates our model of the display ad problem as simultaneously
maximizing weight and cardinality.
We study this multi-objective online allocation problem by providing bicrite-
ria competitive algorithms and showing the tightness of our algorithms by proving
hardness results. We [60] present a bicriteria (p(1 − 1/e1/p), (1 − p)(1 − 1/e1−p))-
approximation algorithm for every p in range (0, 1) where the two approximation
factors are for the weight and cardinality objectives. We provide hardness results
that show the exact tightness of our algorithm at three main points: the starting
13
point (0, 1− 1/e), the middle point (0.43, 0.43), and the final point (1− 1/e, 0). Our
hardness results also show that the gap between our algorithm approximation factors
and the pareto optimal curve is at most 9% at other points.
To achieve efficient algorithms that focus on both objectives, we use two primal
dual subroutines that are responsible for weight and cardinality of the allocation
separately. We allow the subroutines to share and exhaust the total capacities of
advertisers separately, and provide an analysis to handle the collisions. To get hard-
ness results, we construct multi-phase instances with exponentially growing weights.
We then formulate linear programs that upper bound the performance of any on-
line algorithm on the synthesized instances. Approximating the linear programs with
constant size factor-revealing LPs yields the appropriate hardness results.
1.3 Submodular Secretary Problem
Another interesting and well-studied online allocation problem is the classic secretary
problem in which a company wants to hire a secretary with a sequence of applicants
arriving one by one. The goal is to hire the maximum value applicant, and the main
assumption that yields non-zero competitive algorithms in this setting is the random
order arrival of applicants. One of the most important generalizations of the secretary
problem is the multiple choice version in which the company wants to hire several
skilled workers (applicants) with different constraints varying from a simple cap on
the number of secretaries to much more complex cases, such as matroid constraints.
We consider the problem of hiring up to a fixed number of applicants to maximize
the performance of the secretarial group [11]. The overlapping skills of applicants can
be modeled by defining the performance function as a submodular function on the set
of applicants. The class of submodular functions spans a variety of value functions
in practical settings. Following we show how this generalized version of classic sec-
retary problem relates to the ad auctions applications. Consider an advertiser that
has limited budget to spend throughout the day. She wants to show her ads to a
subset of online users (e.g. associated with searched keywords in a search engine) to
14
maximize her own value while respecting her own budget. The budget constraint of
the advertiser could be modelled with a knapsack constraint when the price of win-
ning each online node is estimated accurately. The value function of subsets of online
nodes is submodular in many applications as it observes the diminishing marginal
value property.
We present constant competitive algorithms for this submodular secretary problem
and provide constant hardness results as well. We then generalize our result to achieve
constant competitive algorithms when we have multiple knapsack constraints. In
the case of multiple matroid constraints, we present poly-logarithmic competitive
algorithms. Finally, we show that Θ(√n) is the best achievable competitve ratio
for the subadditive secretary problem which models almost all practical situations
including this class of submodular secretary problems.
15
Chapter 2
Simultaneous Algorithms for
Budgeted Allocation Problem
Online bipartite matching is a fundamental optimization problem with many appli-
cations in online resource allocation, especially the online allocation of ads on the
Internet. In this problem, we are given a bipartite graph G = (X, Y,E) with a set of
fixed nodes (or bins) Y , a set of online nodes (or balls) X, and a set E of edges between
them. Any fixed node (or bin) yj ∈ Y is associated with a total weighted capacity
(or budget) cj. Online nodes (or balls) xi ∈ X arrive online along with their incident
edges (xi, yj) ∈ E(G) and their weights wi,j. Upon the arrival of a node xi ∈ X, the
algorithm can assign xi to at most one bin yj ∈ Y where (xi, yj) ∈ E(G) and the
total weight of nodes assigned to yj does not exceed cj. The goal is to maximize the
total weight of the allocation. This problem is known as the AdWords problem, and
it has been studied under the assumption thatmaxi,j wi,jminj cj
→ 0, in [65, 18, 23].
Under the most basic online model, known as the adversarial model, the online
algorithm does not know anything about the xi’s or E(G) beforehand. In this model,
the seminal result of Karp, Vazirani and Vazirani [56] gives an optimal online 1− 1e-
competitive algorithm to maximize the size of the matching for unweighted graphs
where wij = 1 for each (xi, yj) ∈ E(G). For weighted graphs, Mehta et al. [65, 18]
presented the first 1 − 1e-approximation algorithm to maximize the total weight of
the allocation for the AdWords problem and this result has been generalized to more
16
general weighted allocation problems [18, 33].
Other than the adversarial model, motivated by applications in online advertising,
various stochastic online models have been proposed for this problem. In such stochas-
tic models, online nodes xi ∈ X arrive in a random order. In other words, given a
random permutation σ ∈ Sn, the ball xσ(t) arrives at time time t for t = 1, 2, . . . , n.
In this case, the seminal result of Devanur and Hayes [23] gives a 1−ε-approximation
for the problem if the number of balls n is a prior information to the algorithm, and
OPTwij≥ O(m logn
ε3), where m := |Y |. This result has also been generalized and improved
in several followup work [34, 2, 76], and its related models like the iid models with
known or unknown distributions [36, 10, 66, 25]1. These stochastic models are partic-
ularly motivated in the context of online ad allocation. In this context, online nodes
correspond to page-views, search queries, or online requests for ads. In these settings,
the incoming traffic of page-views may be predicted with a reasonable precision using
a vast amount of historical data.
All these stochastic models and their algorithms are useful only if the incoming
traffic of online nodes (e.g. page-views) can be predicted with a reasonably good
precision. In other words, such algorithms may rely heavily on a precise forecast of
the online traffic patterns, and may not react quickly to sudden changes in the traffic.
In fact, the slow reaction to such traffic spikes impose a serious limitation in the real-
world use of stochastic algorithms in practical applications. This is a common issue in
applying stochastic optimization techniques for online resource allocation problems
(see e.g., [79]). Various methodologies such as robust or control-based stochastic
optimization [13, 14, 79, 74] have been applied to alleviate this drawback. In this
chapter, we study this problem from a more idealistic perspective and aim to design
algorithms that simultaneously achieve optimal approximation ratios for both the
adversarial and stochastic models. Our goal is to design algorithms that achieve
good performance ratios both in the worst case and in the average case.
Our Contributions and Techniques. In this chapter, we study simultaneous
approximation algorithms for the adversarial and stochastic models for the online
1In the iid stochastic models, online nodes are drawn iid from a known or an unknown distribution.
17
budgeted allocation problem. Our goal is to design algorithms that achieve a com-
petitive ratio strictly better than 1−1/e on average, while preserving a nearly optimal
worst case competitive ratio. Ideally, we want to achieve the best of both worlds, i.e,
to design an algorithm with the optimal competitive ratio in both the adversarial
and random arrival models. Toward this goal, we show that this can be achieved for
unweighted graphs, but not for weighted graphs. Nevertheless, we present improved
approximation algorithms for weighted graphs.
For weighted graphs we prove that no algorithm can simultaneously achieve nearly
optimal competitive ratios on both the adversarial and random arrival models. In
particular, we show that no online algorithm that achieve an approximation factor of
1 − 1e
for the worst-case inputs may achieve an average approximation factor better
than 97.6% for the random inputs (See Corollary 2.5.3). More generally, we show that
any algorithm achieving an approximation factor of 1−ε in the stochastic model may
not achieve a competitive ratio better than 4√ε in the adversarial model (See Theorem
2.5.1). In light of this hardness result, we aim to design algorithms with improved
approximation ratios in the random arrival model while preserving the competitive
ratio of 1 − 1e
in the worst case. To this end, we show an almost tight analysis of
the algorithm proposed in [65] in the random arrival model. In particular, we show
its competitive ratio is at least 0.76, and is no more than 0.81 (See Theorem 2.3.1,
and Lemma 2.5.5). Combining this with the worst-case ratio analysis of Mehta et
al. [65] we obtain an algorithm with the competitive ratio of 0.76 for the random
arrival model, while having a 1 − 1e
ratio in the worst case. It is worth noting that
unlike the result of [23] we do not assume any prior knowledge of the number of balls
is given to the algorithm.
On the other hand, for unweighted graphs, under some mild assumptions, we
show a generalization an algorithm in [54] achieves a competitive ratio of 1 − ε in
the random arrival model (See Theorem 2.4.1). Combining this with the worst-
case ratio analysis of [54, 65], we obtain an algorithm with the competitive ratio of
1− ε in the random arrival model, while preserving the optimal competitive ratio of
1 − 1e
in the adversarial model. Previously, a similar result was known for a more
18
restricted stochastic model where all bins have equal capacities [68]. For the case
of small degrees, an upper bound of 0.82 is known for the approximation ratio of
any algorithm for the online stochastic matching problem (even for the under the iid
model with known distributions) [66].
Our proofs consist of three main steps: (i) the main technique is to define an
appropriate potential function as an indefinite integral of a scoring function, and in-
terpret the online algorithms as a greedy algorithm acting to improve these potential
functions by optimizing the corresponding scoring functions (see Section 2.2); These
potential functions may prove useful elsewhere; (ii) An important component of the
proof is to write a factor-revealing mathematical program based on the potential
function and its changes, and finally (iii) the last part of the proofs involve changing
the factor-revealing programs to a constant-size LP and solve it using a solver (in the
weighted case), or analyzing the mathematical program explicitly using an interme-
diary algorithm with an oracle access to the optimum (in the unweighted case). The
third step of the proof in the weighted case is inspired by the technique employed
by Mahdian and Yan [64] for unweighted graphs, however, the set of mathematical
programs we used are quite different from theirs.
All of our results hold under two mild assumptions: (i) large capacities (i.e.,
maxi,j wi,jminj cj
→ 0), and (ii) a mild lower bound on the value of OPT: the aggregate
sum of the largest weight ball assigned to each bin by the optimum is much smaller
than OPT, i.e.,∑
j maxi:opt(i)=j wi,j OPT. Both of these assumptions are valid in
real-world applications of this problem in online advertising. The first assumption
also appears in the AdWords problem, and the second assumption aims to get rid of
some degenerate cases in which the optimum solution is very small.
2.1 Notations
Let G(X, Y,E) be a (weighted) bipartite graph, where X := x1, . . . , xn is the set of
online nodes (or balls), and Y := y1, . . . , ym is the set of fixed nodes (or bins). For
each pair of nodes xi, yj, wi,j represents the weight of edge (xi, yj). Each online node
19
yj is associated with a weighted capacity (or budget) cj > 0. The online matching
problem is as follows: first a permutation σ ∈ Sn is chosen (the distribution may be
chosen according to any unknown distribution): at times t = 1, 2, . . . , n, the ball xσ(t)
arrives and its incident edges are revealed to the algorithm. The algorithm can assign
this ball to at most one of the bins that are adjacent to it. The total weight of balls
assigned to each bin yj may not exceed its weighted capacity cj. The objective is to
maximize the weight of the final matching.
Given the graph G, the optimum offline solution is the maximum weighted bipar-
tite matching in G respecting the weighted capacities, i.e, the total weight of balls
assigned to to a bin yj may not exceed cj. For each ball xi, let opt(i) denote the index
of the bin that xi is being matched to in the optimum solution, and alg(i) be the
index of the bin that xi is matched to in the algorithm. Also for each node yj ∈ Y ,
let oj be the weighted degree of yj in the optimum solution. Observe that for each
j, we have 0 ≤ oj ≤ cj. By definition, we have the size of the optimum solution
is OPT =∑
j oj. Throughout the chapter, we use OPT as the total weight of the
optimal solution, and ALG as the total weight of the output of the online algorithm.
Throughout this chapter, we make the assumption that the weights of the edges
are small compared to the capacities, i.e., maxi,j wi,j is small compared to mini cj.
Also we assume that the aggregate sum of the largest weight ball assigned to each
bin by the optimum is much smaller than OPT i.e.,∑
j maxi:opt(i)=j wi,j OPT. In
particular, let
γ ≥ max
maxi,j
wi,jcj,
√∑j maxi:opt(i)=j wi,j
OPT
(2.1)
the guarantees of our algorithm are provided for the case when γ → 0. For jus-
tifications behind these mild assumptions, see the discussion in the Introduction.
Throughout this chapter, wlog we assume that the optimum matches all of the balls,
otherwise we can throw out the unmatched ball and it can only make the competitive
ratio of the algorithm worse.
20
2.2 Main Ideas
In this section, we describe the main ideas of the proof. We start by defining the
algorithms as deterministic greedy algorithms optimizing specific scoring functions.
We also define a concave potential function as an indefinite integral of the scoring
function, and show that a “good” greedy algorithm must try to maximize the potential
function. In section 2.2.1, we show that if σ is chosen uniformly at random, then we
can lower-bound the increase of the potential in an ε fraction of process; finally in
section 2.2.2 we write a factor-revealing mathematical program based on the potential
function and its changes.
We consider a class of deterministic greedy algorithms that assign each incoming
ball xσ(t) based on a “scoring function” defined over the bins. Roughly speaking,
the scoring function characterizes the “quality” of a bin, and a larger score implies
a better-quality bin. In this chapter, we assume that the score is independent of
the particular labeling of the bins, and it is a non-negative, non-increasing function
of the amount that is saturated so far (roughly speaking, these algorithms try to
prevent over-saturating a bin when the rest are almost empty). Throughout this
chapter, we assume that the scoring function and its derivative are bounded (i.e.,
|f ′(.)|, |f(.)| ≤ 1). However, all of our arguments in this section can also be applied
to the more general scoring functions that may even depend on the overall capacity
ci of the bins. At a particular time t, let rj(t) represent the fraction of the capacity
of the bin yj that is saturated so far. Let f(rj(t)) be the score of bin yj at time t.
When the ball xσ(t+1) arrives, the greedy algorithm simply computes the score of all
of the bins and assigns xσ(t+1) to the bin yj maximizing the product of wσ(t+1),j and
f(rj(t)).
Kalyanasundaram, and Pruhs [54] designed the algorithm Balance using the scor-
ing function fu(rj(t)) := 1−rj(t) (i.e., the algorithm simply assigns an in-coming ball
to the neighbor with the smallest ratio if it is less than 1, and drops the ball otherwise).
They show that for any unweighted graph G, it achieves a 1− 1/e competitive ratio
against any adversarially chosen permutation σ. Mehta et al. [65] generalized this al-
21
gorithm to weighted graphs by defining the scoring function fw(rj(t)) = (1−e1−rj(t)).
Their algorithm, denoted by Weighted-Balance , achieves a competitive ratio of 1−1/e
for the AdWords problem in the adversarial model. We note that both of the algo-
rithms never over-saturate bins (i.e., 0 ≤ rj(t) ≤ 1). Other scoring functions have also
been considered for other variants of the problem (see e.g. [63, 33]). Intuitively, these
scoring functions are chosen to ensure that the algorithm assigns the balls as close
to opt(xσ(t)) as possible. When the permutation is chosen adversarially, any scoring
function would fail to perfectly monitor the optimum assignment (as discussed be-
fore, no online algorithm can achieve a competitive ratio better than 1 − 1/e in the
adversarial model). However, we hope that when σ is chosen uniformly at random,
for any adversarially chosen graph G, the algorithm can almost capture the optimum
assignment. In the following we try to formalize this observation.
We measure the performance of the algorithm at time t by assigning a potential
function that in some sense compares the quality of the overall decisions of the al-
gorithm w.r.t. the optimum. Assuming the optimum solution saturates all of the
bins (i.e., cj = oj), the potential function achieves its maximum at the end of the
algorithm if the balls are assigned exactly according to the optimum. The closer the
value of the potential function to the optimum means a better assignment of the balls.
We define the potential function as the weighted sum of the indefinite integral of the
scoring functions of the bins chosen by the algorithm:
φ(t) :=∑j
cj
∫ rj(t)
r=0
f(r)dr =∑j
cjF (rj(t)).
In particular, we use the following potential function for the Balance and the
Weighted-Balance algorithms respectively:
φu(t) : = −1
2
∑j
cj(1− rj(t))2 (2.2)
φw(t) : =∑j
cj(rj − erj(t)−1). (2.3)
Observe that since the scoring function is a non-increasing function of the ratios, its
22
antiderivative F (.) will be a concave function of the ratios. Moreover, since it is always
non-negative the value of the potential function never decreases in the running time
of the algorithm. By this definition the greedy algorithm can be seen as an online
gradient descent algorithm which tries to maximize a concave potential function; for
each arriving ball xσ(t), it assigns the ball to the bin that makes the largest local
increase in the function.
To analyze the performance of the algorithm we lower-bound the increase in the
value of the potential function based on the optimum matching. This allows us to
show that the final value of the potential function achieved by the algorithm is close
to its value in the optimum, thus bound the competitive ratio of the algorithm. In the
next section, we use the fact that σ is chosen randomly to lower-bound the increase in
εn steps. Finally, in section 2.2.2 we write a factor-revealing mathematical program
to compute the competitive ratio of the greedy algorithm.
2.2.1 Lower bounding the Increase in the Potential Function
In this part, we use the randomness defined on the permutation σ to argue that with
high probability the value of the potential function must have a significant increase
during the run of the algorithm. We define a particular event E corresponding to
event that the arrival process of the balls is approximately close to its expectation.
To show that E occurs with high probability, we only consider the distribution of
arriving balls at 1/ε equally distance times; as a result we can mainly monitor the
amount of increase in the potential function at these time intervals. For a carefully
chosen 0 < ε < 1, we divide the process into 1/ε slabs such that the kth slab includes
the [knε + 1, (k + 1)nε] balls. Assuming σ is chosen uniformly at random, we show
a concentration bound on the weight of the balls arriving in the kth slab. Using that
we lower bound φ((k + 1)nε)− φ(knε) in Lemma 2.2.2.
First we use the randomness to determine the weight of the balls arriving in the kth
slab. Let Ii,k be the indicator random variable indicating that the ith ball will arrive
in the kth slab. Observe that for any k, the indicators Ii,k are negatively correlated:
knowing that Ii,k = 1 can only decrease the probability of the occurrence of the other
23
balls in the kth slab (i.e., P [Ii,k|Ii′,k = 1] ≤ P [Ii,k]). Define Nj,k :=∑
i:opt(i)=j wi,jIi,k
as the sum of the weight of the balls that are matched to the jth bin in the optimum
and arrive in the kth slab. It is easy to see that Eσ [Nj,k] = ε · oj, moreover, since it is
a linear combination of negatively correlated random variables it will be concentrated
around its mean. Define h(k) :=∑
j |Nj,k − εoj|. The following Lemma shows that
h(k) is very close to zero for all time slabs k with high probability. Intuitively, this
implies that with high probability the balls from each slab are assigned similarly in
the optimum solution.
Lemma 2.2.1. Let h(k) :=∑
j |Nj,k − εoj|. Then Pσ
[∀k, h(k) ≤ 5γ
εδOPT
]≥ 1− δ.
Proof. It suffices to upper-bound P[h(k) > 5γ
εδOPT
]≤ δε; the lemma can then be
proved by a simple application of the union bound. First we use Azuma-Hoeffding
concentration bound to compute E [|Nj,k − εoj|]; then we simply apply the Markov
inequality to upper-bound h(k).
Let Wj :=√
2∑
i:opt(i)=j w2i,j, for any j, k, we show E [|Nj,k − εoj|] ≤ 3Wj. Since
Nj,k is a linear combination of negatively correlated random variables Ii,k for opt(i) =
j, and E [Nj,k] = ε·oj by a generalization of the Azuma Hoeffding bound to negatively
correlated random variables [70] we have
E [|Nj,k − ε · oj|] ≤ Wj
∞∑l=0
P [|Nj,k − E [Nj,k] | ≥ l ·Wj]
≤ Wj
1 + 2∞∑l=1
e−
l2W2j
2∑i:opt(i)=j w
2i,opt(i)
≤ Wj
(1 + 2
∞∑l=1
e−l2
)≤ 3Wj. (2.4)
Let wmax(j) := maxi:opt(i)=j wi,j. Since W 2j as twice the sum of the square of the
weights assigned to the jth bin is a convex function we can write Wj ≤√
2wmax(j)oj.
Therefore, by the linearity of expectation we have
E [h(k)] ≤∑j
3√
2wmax(j)oj ≤ 5∑j
wmax(j)/γ + γoj2
≤ 5 1
2γ
∑j
wmax(j) +γ
2OPT ≤ 5γOPT,
24
where the last inequality follows from assumption (2.1). Since h(k) is a non-negative
random variable, by the Markov inequality we get P[h(k) > 5γ
εδOPT
]≤ δε. The
lemma simply follows by applying this inequality for all k ∈ 0, . . . , 1/ε and using
the union bound.
Let E be the event that ∀k, h(k) ≤ 5γεδ
OPT. The next lemma shows that condi-
tioned on E , one can lower-bound the increase in the potential function in any slab
(i.e., φ((k + 1)nε)− φ(knε) for any 0 ≤ k < 1/ε):
Lemma 2.2.2. Conditioned on E, for any 0 ≤ k < 1/ε, t0 = knε, and t1 = (k+1)nε
we have
φ(t1)− φ(t0) ≥ ε∑j
ojf(rj(t1)) −6√γ
εδOPT.
Proof. First we simply compute the increase of the potential function at time t + 1,
for some t0 ≤ t < t1. Then, we lower-bound the increase using the monotonicity
of the scoring function f(.). Finally, we condition on E and lower-bound the final
expression in terms of OPT. Let σ(t + 1) = i, and assume the algorithm assigns xi
to the jth bin (i.e., alg(i) = j); since the algorithm maximizes wi,jf(rj(t)) we can
lower-bound the increase of the potential function based on the optimum. First using
the mean value theorem of the calculus we have:
cj
F (rj(t) +
wi,jcj
)− F (rj(t)
= cj
wi,jcjf(rj(t)) +
1
2
(wi,jcj
)2
f ′(r∗)
,
for some r∗ ∈ [rj(t), rj(t) + wi,j/cj]. Since the derivative of f(.) is bounded (i.e.,
|f ′(r)| ≤ 1 for all r ∈ [0, 1]), we get
φ(t+ 1)− φ(t) = cj
F (rj(t) +
wi,jcj
)− F (rj(t)
≥ wi,opt(i)f(ropt(i)(t))− wi,j
wi,jcj
≥ wi,opt(i)f(ropt(i)(t1))− γwi,j, (2.5)
where the first inequality follows by the greedy decision chosen by the algorithm
wi,jf(rj(t)) ≥ wi,opt(i)f(ropt(i)(t)), and the last inequality follows by assumption (2.1).
Consider a single run of the algorithm; wlog we assume that ALG ≤ OPT. We
25
can monitor the amount of increase in the potential function in the kth slab as follows:
φ(t1)− φ(t0) ≥t1−1∑t=t0
wσ(t),opt(σ(t))f(ropt(σ(t))(t))− γwσ(t),alg(σ(t))
≥
∑j
∑t0≤t<t1,opt(σ(t))=j
f(rj(t1))wσ(t),j − γOPT
=∑j
f(rj(t1))Nj,k − γOPT
where the second inequality follows by the fact that f(.) is a non-increasing function
of the ratio, and∑
t0≤t<t1 wσ(t),alg(σ(t)) ≤ ALG ≤ OPT, and the equality follows from
the definition of Nj,k. By lemma 2.2.1 we know Nj,k is highly concentrated around
ε · oj. Conditioned on E , we have h(k) ≤ 5γεδOPT , thus:
φ(t1)− φ(t0) ≥ ε∑j
f(rj(t1))oj −∑j
|Nj,k − εoj| − γOPT
≥ ε∑j
f(rj(t1))oj − h(k)− γOPT ≥ ε∑j
f(rj(t1))oj −6γ
εδOPT
where the first inequality follows by the assumption |f(.)| ≤ 1.
2.2.2 Description of the Factor-Revealing Mathematical Pro-
gram
In this section we propose a factor-revealing mathematical program that lower-bounds
the competitive ratio of the algorithms Balance and Weighted-Balance . In sections
2.3, and 2.4 we derive a relaxation of the program and analyze that relaxation. In-
terestingly, the main non-trivial constraints follows from the lower-bounds obtained
for the amount of increase in the potential function in Lemma 2.2.2.
The details of the program is described in MP(1). It is worth noting that the last
inequality in this program follows from the monotonicity property of the ratios. In
other words, we assume the ratio function rj(t) is a monotonically increasing function
w.r.t. to t.
26
MP(1)minimize 1
OPT
∑j minrj(n), 1cj∑
j cjF (rj(t)) = φ(t) ∀t ∈ [n],
ε∑
j ojf(rj((k + 1)nε))− 6γεδ
OPT ≤ φ((k + 1)nε)− φ(knε) ∀k ∈ [1ε− 1],
oj ≤ cj ∀j ∈ [m],∑j oj = OPT,
rj(t) ≤ rj(t+ 1) ∀j, t ∈ [n− 1].
The following proposition summarizes our arguments and shows that the program
MP(1) is a relaxation for any deterministic greedy algorithm that works based on
a scoring function. It is worth noting that the whole argument still follows when
the scoring function is not necessarily non-negative; we state the proposition in this
general form.
Proposition 2.2.3. Let f be any non-increasing, scoring function of the ratios rj(t)
of the bins such that |f(r)|, |f ′(r)| ≤ 1 for the range of ratios that may be encountered
in the running time of the algorithm. For any (weighted) graph G = (X, Y ), and
ε > 0, with probability at least 1−δ, MP(1) is a factor-revealing mathematical program
for the greedy deterministic algorithm that uses scoring function f(.).
Since the potential function F (.) is a concave function, this program may not
be solvable in polynomial time. In section 2.4, we show that after adding a new
constraint it is possible to analyze it analytically for the unweighted graphs. To deal
with this issue for the weighted graphs, we write a constant-size LP relaxation of the
program that lower-bounds the optimum solution (after losing a small error). Finally,
we solve the constant-size LP by an LP solver, and thus obtain a nearly tight bound
for the competitive ratio of the Weighted-Balance (see section 2.3 for more details).
In the rest of this section, we write a simpler mathematical program that will be
used later in section 2.3 for analyzing the Weighted-Balance algorithm. In particular,
we simplify the critical constraint that measures the increase in the potential function
by further removing the term −6γεδ
OPT. Moreover, since Weighted-Balance never
over-saturates bins we can also add the constraint rj(n) ≤ 1 to both MP(1) and
MP(2) and still have a relaxation of Weighted-Balance .
27
MP(2) minimize∑
j rj(n)cjs.t.
∑j cj(rj(t)− erj(t)−1) = φ(t) ∀t ∈ [n],
ε∑
j oj(1− erj((k+1)nε)−1) ≤ φ((k + 1)nε)− φ(knε) ∀k ∈ [1ε],
oj ≤ cj ∀j ∈ [m],∑j oj = 1.
rj(t) ≤ rj(t+ 1) ∀j, t ∈ [n− 1],rj(n) ≤ 1 ∀j ∈ [m].
In the next Lemma we show that the optimum value of MP(2) is at least (1−√
12γε2δ
)
of MP(1):
Lemma 2.2.4. For any weighted graph G, we have MP(1) ≥ (1−α) min1,MP(2),
where α :=√
12γε2δ
.
Proof. Wlog we can replace OPT = 1 in MP(1). Let s1 := rj(t), cj, oj, φ(t) be a
feasible solution of the MP(1). If∑
j rj(n)cj ≥ (1 − α) we are done; otherwise we
construct a feasible solution s2 of the MP(2) such that the value of s1 is at least
(1−α) of the value of s2. Then the lemma simply follows from the fact that the cost
of the value of the optimum solution of MP(1) is at least of (1 − α) of the value of
the optimum of MP(2).
Define s2 := rj(t), cj/(1 − α), oj, φ(t)/(1 − α). Trivially, s2 satisfies all except
(possibly) the second constraint of MP(2). Moreover, the value of s1 is (1− α) times
the value of s2. It remains to prove the feasibility of the second constraint of MP(2),
i.e.,
ε(1− α)∑j
oj(1− erj((k+1)nε)−1) ≤ φ((k + 1)nε)− φ(knε),
for all k ∈ [1/ε]. Since s1 is a feasible solution of MP(1) we have
φ((k + 1)nε)− φ(knε) ≥ ε∑j
oj(1− erj((k+1)nε)−1)− ε
2α2
≥ ε∑j
oj(1− erj((k+1)nε)−1)
1−
ε2α2
ε2
∑j oj(1− rj((k + 1)nε)
, (2.6)
where the last inequality follows from the assumption that 0 ≤ rj(t) ≤ 1, and the fact
that 1− ex−1 ≥ 12(1− x) for x ∈ [0, 1]. On the other hand, since
∑j rj(n)cj < 1− α,
28
we can write: ∑j
oj(1− rj((k + 1)nε)) ≥ 1−∑j
cjrj(n) ≥ α
The lemma simply follows from putting the above inequality together with equation
(2.6).
2.3 The Competitive Ratio of Weighted-Balance
In this section, we lower-bound the competitive ratio of the WEIGHED-BALANCE
algorithm in the random arrival model. More specifically, we prove the following
theorem:
Theorem 2.3.1. For any weighted graph G = (X, Y ), and
δ > 0, with probability 1 − δ, the competitive ratio of the
Weighted-Balance algorithm in the random arrival model is at least
0.76(1−O(√γ/δ)).
To prove the bound in this theorem, we write a constant-size linear programming
relaxation of the problem based on MP(2) and solve the problem by an LP solver.
The main two difficulties with solving program MP(2) are as follows: first, as we
discussed in section 2.2.2, MP(2) is not a convex program; second, the size of the
program (i.e., the number of variables and constraints) is a function of the size of the
graph G. The main idea follows from a simple observation that the main inequalities
in MP(2), those lower-bounding the increase in the potential function, are indeed
lower-bounding the increase in the potential function only at constant (i.e., 1/ε)
number of times. Hence, we do not need to keep track of the ratios and the potential
function for all t ∈ [n]; instead it suffices to monitor these values at 1/ε critical times
(i.e., at times knε for k ∈ [1/ε]), for a constant ε. Even in those critical times it
suffices to approximately monitor the ratios of the bins by discretizing the ratios into
1/ε slabs.
For any integers 0 ≤ i < 1/ε, 0 ≤ k ≤ 1/ε, let ci,k be the sum of the capacities of
the bins of ratio rj(knε) ∈ [iε, (i + 1)ε), and oi,k be the sum of the weighted degree
29
of the bins of ratio rj(knε) ∈ [iε, (i+ 1)ε) in the optimum solution, i.e.,
ci,k :=∑
j:rj(knε)∈[iε,(i+1)ε)
cj, oi,k :=∑
j:rj(knε)∈[iε,(i+1)ε)
oj. (2.7)
Now we are ready to describe the constant-size LP relaxation of MP(2). We write the
LP relaxation in terms of the new variables ci,k, oi,k. In particular, instead of writing
the constraints in terms of the actual ratios of the bins, we round down (or round
up) the ratios to the nearest multiple of ε such that the constraint remains satisfied.
The details are described in LP(1).
LP(1) minimize 11−1/e
φ(1
ε)−
∑1/ε−1i=0 ci,k(iε/e− eiε−1)
s.t.
∑1/ε−1i=0 ci,k(iε− eiε−1) ≤ φ(k) ∀k ∈ [1
ε]∑1/ε−1
i=0 εoi,k+1(1− e(i+1)ε−1) ≥ φ(k + 1)− φ(k) ∀k ∈ [1ε− 1]
ci,k ≥ oi,k ∀i ∈ [1ε− 1], k ∈ [1
ε]∑1/ε−1
i=0 oi,k = 1 ∀k ∈ [1ε] :∑1/ε−1
l=i cl,k+1 ≥∑1/ε−1
l=i cl,k ∀i ∈ [1ε− 1], k ∈ [1
ε− 1]
In the next Lemma, we show that the LP(1) is a linear programming relaxation
of the program MP(2):
Lemma 2.3.2. The optimum value of LP(1) lower-bounds the optimum value of
MP(2).
Proof. We show that for any feasible solution s := rj(t), cj, oj, φ(t) of MP(2) we can
construct a feasible solution s′ = c′i,k, o′i,k, φ′(k) for LP(1) with a smaller objective
value. In particular, we construct s′ simply by using equation (2.7), and letting
φ′(k) := φ(knε). First we show that all constraints of LP(1) are satisfied by s′, then
we show that the value of LP(1) for s′ is smaller than the value of MP(2) for s.
The first equation of LP(1) simply follows from rounding down the ratios in the
first equation of MP(2) to the nearest multiple of ε. The equation remains satisfied by
the fact that the potential function φ(.) is increasing in the ratios (i.e., Fw(r) = r−er−1
is increasing in r ∈ [0, 1]). Similarly, the second equation of LP(1) follows from
rounding up the ratios in the second equation of MP(2), and noting that the scoring
function is decreasing in the ratios (i.e., fw(r) = 1− er−1 is decreasing for r ∈ [0, 1]).
30
The third and fourth equations can be derived from the corresponding equations in
MP(2). Finally, the last equation follows from the monotonicity property of the ratios
(i.e., rj(t) is a non-decreasing function of t).
It remains to compare the values of the two solutions s, and s′. We have
1
1− 1/e
φ′(1
ε)−
1/ε−1∑i=0
c′i,1/ε(iε/e− eiε−1
) ≤
1
1− 1/e
φ(n)−
∑j
cj(rj(n)
e− erj(n)−1)
=
∑j
cjrj(n),
where the inequality follows from the fact that r/e− er−1 is a decreasing function for
r ∈ [0, 1], and the last inequality simply follows from the definition of φw(.) (i.e., the
first equation of MP(2)).
Now we are ready to prove Theorem 2.3.1:
Proof of Theorem 2.3.1. By Proposition 2.2.3, for any ε > 0, with probability
1 − δ the competitive ratio of Weighted-Balance is lower-bounded by the optimum
of MP(1). On the other hand, by Lemma 2.2.4 the optimum solution of MP(1) is
at least (1−√
12γε2δ
) of the optimum solution of MP(2). Finally, by Lemma 2.3.2 the
optimum solution of MP(2) is at least the optimum of LP(1). Hence, with probability
1−δ the competitive ratio of Weighted-Balance is at least (1−√
12γε2δ
) of the optimum
of LP(1).
The constant-size linear program LP(1) can be solved numerically for any value
of ε > 0. By solving this LP using an LP solver, we can show that for ε = 1/250 the
optimum solution is greater than 0.76. This implies that with probability 1 − δ the
competitive ratio of Weighted-Balance is at least 0.76(1−O(√γ/δ)).
Remark 2.3.3. We also would like to remark that the optimum solution of LP(1)
beats the 1 − 1/e factor even for ε = 1/8; roughly speaking this implies that even if
the permutation σ is almost random, in the sense that each 1/8 of the input almost
has the same distribution, then Weighted-Balance beats the 1− 1/e factor.
31
2.4 The Competitive Ratio of Balance
In this section we show that for any unweighted graph G, under some mild assump-
tions, the competitive ratio of Balance approaches 1 in the random arrival model.
Theorem 2.4.1. For any unweighted bipartite graph G = (X, Y,E), and δ > 0, with
probability 1 − δ the competitive ratio of Balance in the random arrival model is at
least 1− β∑j cj
OPT, where β := 3(γ/δ)1/6.
First we assume that our instance is all-saturated meaning that the optimum
solution saturates all of the bins (i.e., cj = oj for all j), and show that the competitive
ratio of the algorithm is at least 1− 3(γ/δ)1/6:
Lemma 2.4.2. For any δ > 0, with probability 1 − δ the competitive ratio of Bal-
ance on all-saturated instances in the random arrival model is at least 1− β.
Then we prove Theorem 2.4.1 via a simple reduction to the all-saturated case.
To prove Lemma 2.4.2, we analyze a slightly different algorithm Balance’ that
always assigns an arriving ball (possibly to an over-saturated bin); this will allow us
to keep track of the number of assigned balls at each step of the process. In particular
we have
∀t ∈ [n] :∑j
cjrj(t) = t, (2.8)
where rj(t) does not necessarily belong to [0, 1]. The latter certainly violates some of
our assumptions in Section 2.2. To avoid the violation, we provide some additional
knowledge of the optimum solution to Balance’ such that the required assumptions
are satisfied, and it achieves exactly the same weight as Balance .
We start by describing Balance’ , then we show that it still is a feasible algorithm
for the potential function framework studied in Section 2.2; in particular we show it
satisfies Proposition 2.2.3. When a ball xi arrives at time t + 1 (i.e., σ(t + 1) = i),
similar to Balance , Balance’ assigns it to a bin maximizing wi,jfu(rj(t)); let j be such
a bin. Unlike Balance if rj(t) ≥ 1 (i.e., all neighbors of xi are saturated), we do not
drop xi; instead Balance’ assigns it to the bin yopt(i).
32
First note that although Balance’ magically knows the optimum assignment of a
ball once all of its neighbors are saturated, it achieves the same weight matching. This
simply follows from the fact that over-saturating bins does not increase our gain, and
does not alter any future decisions of the algorithm. Next we use Proposition 2.2.3
to show that MP(1) is indeed a mathematical programming relaxation for Balance’ .
By Proposition 2.2.3, we just need to verify that |fu(.)|, |f ′u(.)| ≤ 1 for all the ratios
we might encounter in the run of Balance’ . Since fu(r) = (1− r), and the ratios are
always non-negative, it is sufficient to show that the ratios are always upper-bounded
by 2. To prove this, we crucially use the fact that Balance’ has access to the optimum
assignment for the balls assigned to the over-saturated bins. Observe that the set
of balls assigned to a bin after it is being saturated, is always a subset of the balls
assigned to it in the optimum assignment. Since the ratio of all bins are at most 1 in
the optimum, they will be upper-bounded by 2 in Balance’ .
The following is a simple mathematical programming relaxation to analyze Bal-
ance’ in the all-saturated instances:
MP(3)minimize
∑j minrj(n), 1cj∑
j cjrj(t) = t t ∈ [n],
ε∑
j cj(1− rj((k + 1)nε))− 6γεδ
OPT ≤ φu((k + 1)nε)− φu(knε) ∀k ∈ [1ε− 1],
Note that the first constraint follows from (2.8), and the second constraint follows
from the second constraint of MP(1), and the fact that cj = oj in the all-saturated
instances.
Now we are ready to prove Lemma 2.4.2:
Proof of Lemma 2.4.2. With probability 1−δ, MP(3) is a mathematical programming
relaxation of Balance’ . First we sum up all 1/ε second constraints of MP(3) to
obtain a lower-bound on φu(n), and we get φu(n) is very close to zero (intuitively,
the algorithm almost manages to optimize the potential function). Then, we simply
apply the Cauchy-Schwarz inequality to φu(n) to bound the loss of Balance’ .
We sum up the second constraint of MP(3) for all k ∈ 0, 1, . . . , 1ε− 1; the RHS
33
telescopes and we obtain:
φu(n)− φu(0) ≥ OPT (1− 6γ
ε2δ)− ε
1/ε−1∑k=0
∑j
cjrj((k + 1)nε)
≥ n(1− 6γ
ε2δ)− ε2n
1/ε−1∑k=0
(k + 1) ≥ n(1
2− ε
2− 6γ
ε2δ)
where the first inequality follows by the assumption that the instance is all-saturated,
and the second inequality follows from applying the first constraint of MP(3) for
t = (k+ 1)nε, and the fact that OPT = n. Since φu(0) = −12
∑j(1− rj(0))2 = −n/2,
we obtain φu(n) ≥ −n( ε2
+ 6γε2δ
).
Observe that only the non-saturated bins incur a loss to the algorithm, i.e.,
Loss(Balance’ ) =∑
rj(n)<1
cj(1− rj(n)).
Using the lower-bound on φu(n) we have
∑rj(n)<1
cj(1− rj(n)) ≤√ ∑
rj(n)<1
cj(1− rj(n))2 ·∑
rj(n)<1
cj
≤√−2φu(n) · n ≤ n
√ε+
12γ
ε2δ,
where the first inequality follows by the Cauchy-Schwarz inequality, and the second
inequality follows from the definition of φu(n). The lemma simply follows from choos-
ing ε = 2(2γ/δ)1/3 in the above inequality.
Next we prove Theorem 2.4.1; we analyze the general instances by a reduction to
all-saturated instances.
Proof of Theorem 2.4.1. Let G = (X, Y ) be an unweighted graph, similar to Lemma
2.4.2 it is sufficient to analyze Balance’ on G. For every bin yj we introduce cj − ojdummy balls that are only adjacent to the jth bin, and let G′ = (X ′, Y ) be the new
instance. First we show that the expected number of non-dummy balls matched by
Balance’ in G′ is at most the expected size of the matching that Balance’ achieves
34
in G. We analyze the performance of Balance’ on G simply using Lemma 2.4.2, and
eliminating the effect of dummies.
Fix a permutation σ ∈ S|X′|; letW ′(σ) be the number of non-dummy balls matched
by Balance’ on σ. Similarly, let W (σ[X]) be the size of the matching obtained on
σ[X] in G, where σ[X] is the projection of σ on X. Using an argument similar to [17,
Lemma 2] (e.g., the monotonicity property), one can show that W ′(σ) ≤ W (σ[X])
for all σ ∈ S|X′|. Hence, to compute the competitive ratio of Balance’ on G, it is
sufficient to upper-bound the expected number of non-dummy balls not-matched by
Balance’ on G′. The latter is certainly not more than the total loss of Balance’ on G′
which is no more than β∑
j cj by Lemma 2.4.2.
2.5 Hardness Results
In this section, we show that there exists a family of weighted graphs G such that
for any ε > 0, any online algorithm that achieves a 1 − ε competitive ratio in the
random arrival model, does not achieve an approximation ratio better than a function
g(ε) in the adversarial model, where g(ε) → 0 as ε → 0. More specifically, we prove
something stronger:
Theorem 2.5.1. For any constants δ, ε > 0, there exists family of weighted bipartite
graphs G = (X, Y ) such that for any (randomized) algorithm that achieves 1 − ε
competitive ratio (in expectation) on at least δ fraction of the permutations σ ∈ S|X|,
does not achieve more than 4√ε (in expectation) for a particularly chosen permutation
in another graph G′.
As a corollary, we can show that any algorithm that achieves the competitive
ratio of 1 − 1/e in the adversarial model can not achieve an approximation factor
better than 0.976 in the random arrival model. Moreover, at the end of this section,
we show that for some family of graphs the Weighted-Balance algorithm does not
achieve an approximation factor better than 0.81 in the random arrival model (see
Lemma 2.5.5 for more details). This implies that our analysis of the competitive ratio
35
of this algorithm is tight up to an additive factor of 5%. We start by presenting the
construction of the hard examples:
Example 2.5.2. Fix a large enough integer l > 0, and let α :=√ε; let Y := y1, y2
with capacities c1 = c2 = l. Let C and D be two types of balls (or online nodes), and
let the set of online nodes X correspond to a set of l copies of C and l/α copies of
D. Each type C ball has a weight of 1 for y1, and a weight of 0 for y2, while a type
D ball has a weight of 1 in y1 and a weight of α in y2.
First of all, observe that the optimum solution achieves a matching of weight 2l
simply by assigning all type C balls to y1, and type D balls to y2. On the other hand,
any algorithm that achieves the competitive ratio of 1−ε in the random arrival model
should match the balls “very similar” to this strategy. However, if the algorithm uses
this strategy, then an adversary may construct an instance by preserving the first
l balls of the input followed by l/α dummy balls. But in this new instance it is
“much better” to assign all of the first l balls to y1. In the following we formalize this
observation.
Proof of Theorem 2.5.1. Let G be the graph constructed in Example 2.5.2, and let
A be a (randomized) algorithm that achieves 1− ε competitive ratio (in expectation)
on at least δ fraction of permutations σ ∈ Sn, where n = l + l/α, for some constant
δ > 0. First we show that there exists a particular permutation σ∗ such that there
are at most lα balls of type C among σ∗(1), . . . , σ∗(l), and algorithm A achieves at
least (1−ε)2l on σ∗. Then we show that the (expected) gain of A from the first l balls
is at most 4l√ε. Finally, we construct a new graph G′ = (X ′, Y ) and a permutation
σ′ such that the first l balls in σ′ is the same as the first l balls of σ∗. This will imply
that A does not achieve a competitive ratio better than 4√ε on G′.
To find σ∗ it is sufficient to show that with probability strictly more than 1 − δ
the number of type C balls among the first l balls of a uniformly random chosen
permutation σ is at most lα. This can be proved simply using the Chernoff-Hoeffding
bound. Let Bi be a Bernoulli random variable indicating that xσ(i) is of type C, for
1 ≤ i ≤ l. Observe that Eσ [Bi] = α1+α
, and these variables are negatively correlated.
36
By a generalization of Chernoff-Hoeffding bound [70] we have
P
[l∑
i=1
Bi > αl
]≤ e−
lα3
6 < δ,
where the last inequality follows by choosing l large enough. Hence, there exists a
permutation σ∗ such that the number of type C balls among its first l balls is at most
lα, and A achieves (1− ε)2l on σ∗.
Next we show that the (expected) gain of A from the first l balls of σ∗ is at most
2l(α + ε/α) = 4l√ε. This simply follows from the observation that any ball of type
D that is assigned to y1 incurs a loss of α. Since the expected loss of the algorithm
is at most 2lε on σ∗, the expected number of type D balls assigned to y1 (in the
whole process) is no more than 2lεα
. We can upper-bound the (expected) gain of the
algorithm from the first l balls by lα+ 2lεα
+ lα, where the first term follows from the
upper-bound on the number of C balls, and the last term follows from the number of
D balls (that may possibly be) assigned to y2.
It remains to construct the adversarial instance G′ together with the permutation
σ′. G′ has the same set of bins, while X ′ is the union of the first l balls of σ∗ with
l/α dummy balls (a dummy ball has zero weight in both of the bins). We construct
σ′ by preserving the first l balls of σ∗, filling the rest with the dummy balls (i.e.,
xσ′(i) = xσ∗(i) for 1 ≤ i ≤ l). First, observe that the optimum solution in G′ achieves
a matching of weight l simply by assigning all of the first l balls to y1. On the other
hand, as we proved the (expected) gain of the algorithm A is no more than 4l√ε on
G′. Therefore, the competitive ratio of A in this adversarial instance is no more than
4√ε.
The following corollary can be proved simply by choosing δ small enough in The-
orem 2.5.1:
Corollary 2.5.3. For any constant ε > 0, any algorithm that achieves a competitive
ratio of 1 − ε in the random arrival model does not achieve strictly better than 4√ε
in the adversarial model. In particular, it implies that any algorithm that achieves
a competitive ratio of 1 − 1e
in the adversarial model does not achieve strictly better
37
than 0.976 in the random order model.
It is also worth noting that Weighted-Balance achieves at least 0.89 competitive
ratio in the random arrival model for Example 2.5.2, and the worst case happens for
α ≈ 0.48. Next we present a family of examples where the Weighted-Balance does
not achieve a factor better than 0.81 in the random arrival model.
Example 2.5.4. Fix a large enough integer n > 0, and α < 1; again let Y := y1, y2
with capacities c1 = n, and c2 = n2. Let X be a union of n identical balls each of
weight 1 for y1 and α for y2.
Lemma 2.5.5. For a sufficiently large n, and a particularly chosen α > 0, the
competitive ratio of the Weighted-Balance in the random arrival model for Example
2.5.4 is no more than 0.81.
Proof. First observe that the optimum solution achieves a matching of weight n sim-
ply by assigning all balls to y1. Intuitively, Weighted-Balance starts with the same
strategy, but after partially saturating y1, it sends the rest to y2 (note that each ball
that is sent to y2 incurs a loss of 1 − α to the algorithm). Recall that r1(n) is the
ratio of y1 at the end of the algorithm. The lemma essentially follows from upper-
bounding r1(n) by 1 + 1/n + ln(1 − α(1 − e1/n−1)). Since the algorithm achieves a
matching of weight exactly r1(n)n + (1 − r1(n))nα, and OPT = n, the competitive
ratio is r1(n) + (1− r1(n))α. By optimizing over α, one can show that the minimum
competitive ratio is no more than 0.81, and it is achieved by choosing α ' 0.55.
It remains to show that r1(n) ≤ 1 + 1/n+ ln(1− α(1− e1/n−1)). Let t be the last
time where a ball is assigned to y1 (i.e., r1(t − 1) + 1/n = r1(t) = r1(n)). Since the
ball at time t is assigned to y1, we have
1 · fw(r1(t− 1)) ≥ α · fw(r2(t− 1)) ≥ α · fw(1
n),
where the last inequality follows by the fact that the ratio of the second bin can
not be more than α · n/c2 < 1/n, and fw(.) is a non-increasing function of the
38
ratios. Using fw(r) = 1 − er−1, and r1(t − 1) + 1/n = r1(n) we obtain that r1(n) ≤
1 + 1/n+ ln(1− α(1− e1/n−1)).
2.6 Related Work
In this section, we discuss the related works that are not mentioned at the beginning
of Chapter 2. For unweighted graphs, it has been recently observed that the Karp-
Vazirani-Vazirani 1− 1e-competitive algorithm for the adversarial model also achieves
an improved approximation ratio of 0.70 in the random arrival model [55, 64]. This
holds even without the assumption of large degrees. It is known that without this
assumption, one cannot achieve an approximation factor better than 0.82 for this
problem (even in the case of iid with known distributions) [66]. This is in contrast
with our result for unweighted graphs with large degrees.
Online budgeted allocation and its generalizations appear in two main categories
of online advertising: AdWords (AW) problem [65, 18, 23], and the Display Ads
Allocation (DA) problem [33, 34, 2, 76]. In both of these problems, the publisher
must assign online page-views (or impressions) to an inventory of ads, optimizing
efficiency or revenue of the allocation while respecting pre-specified contracts. In the
DA problem, given a set of m advertisers with a set Sj of eligible impressions and
demand of at most n(j) impressions, the publisher must allocate a set of n impressions
that arrive online. Each impression i has value wij ≥ 0 for advertiser j. The goal
of the publisher is to assign each impression to one advertiser maximizing the value
of all the assigned impressions. The adversarial online DA problem has been studied
in [33] in which the authors present a 1− 1e-competitive algorithm for the DA problem
under a free disposal assumption for graphs of large degrees. This result generalizes
the 1− 1e-approximation algorithm by Mehta et al [65] and by Buchbinder et. al. [18].
Following a training-based dual algorithm by Devanur and Hayes [23] for the AW
problem, training-based (1 − ε)-competitive algorithms have been developed for the
DA problem and its generalization to packing linear programs [34, 76, 2] including
the DA problem. These papers develop a (1 − ε)-competitive algorithm for online
39
stochastic packing problems in which OPTwij≥ O(m logn
ε3) (or OPT
wij≥ O(m logn
ε2) applying
the technique of [2]) and the demand of each advertiser is large, in the random-order
and the i.i.d model. Although studying a similar set of problems, none of the above
papers study the simultaneous approximations for adversarial and stochastic models,
and the dual-based 1 − ε-competitive algorithms for the stochastic variants do not
provide a bounded competitive ratio in the adversarial model.
Dealing with traffic spikes and inaccuracy in forecasting the traffic patterns is a
central issue in operations research and stochastic optimization. Various method-
ologies such as robust or control-based stochastic optimization [13, 14, 79, 74] have
been proposed. These techniques either try to deal with a larger family of stochastic
models at once [13, 14, 79], try to handle a large class of demand matrices at the
same time [79, 4, 6], or aim to design asymptotically optimal algorithms that react
more adaptively to traffic spikes [74]. These methods have been applied in particular
for traffic engineering [79] and inter-domain routing [4, 6]. Although dealing with
similar issues, our approach and results are quite different from the approaches taken
in these papers. Finally, an interesting related model for combining stochastic and
online solutions for the Adwords problem is considered in [63], however their approach
does not give an improved approximation algorithm for the iid model.
40
Chapter 3
Bicriteria Online Matching:
Maximizing Weight and
Cardinality
In the past decade, there has been much progress in designing better algorithms for
online matching problems. This line of research has been inspired by interesting com-
binatorial techniques that are applicable in this setting, and by online ad allocation
problems. For example, the display advertising problem has been modeled as maxi-
mizing the weight of an online matching instance [35, 34, 25, 15, 24]. While weight is
indeed important, this model ignores the fact that cardinality of the matching is also
crucial in the display ad application. This example illustrates the fact that in many
real applications of online allocation, one needs to optimize multiple objective func-
tions, though most of the previous work in this area deals with only a single objective
function. On the other hand, there is a large body of work exploring offline multi-
objective optimization in the approximation algorithms literature. In this chapter, we
focus on simultaneously maximizing online two objectives which have been studied
extensively in matching problems: cardinality and weight. Besides being a natural
mathematical problem, this is motivated by online display advertising applications.
Applications in Display Advertising. In online display advertising, advertisers
typically purchase bundles of millions of display ad impressions from web publishers.
41
Display ad serving systems that assign ads to pages on behalf of publishers must
satisfy the contracts with advertisers, respecting targeting criteria and delivery goals.
Modulo this, publishers try to allocate ads intelligently to maximize overall quality
(measured, for example, by clicks), and therefore a desirable property of an ad serv-
ing system is to maximize this quality while satisfying the contracts to deliver the
purchased number n(a) impressions to advertiser a. This has been modeled in the
literature (e.g., [35, 2, 76, 25, 15, 24]) as an online allocation problem, where quality
is represented by edge weights, and contracts are enforced by overall delivery goals:
While trying to maximize the weight of the allocation, the ad serving systems should
deliver n(a) impressions to advertiser a. However, online algorithms with adversarial
input cannot guarantee the delivery of n(a) impressions, and hence the goals n(a)
were previously modeled as upper bounds. But maximizing the cardinality subject to
these upper bounds is identical to delivering as close to the targets as possible. This
motivates our model of the display ad problem as simultaneously maximizing weight
and cardinality.
Problem Formulation. More specifically, we study the following bicriteria online
matching problem: consider a set of bins (also referred to as fixed nodes, or ads) A
with capacity constraints n(a) > 0, and a set of online items (referred to as online
nodes, or impressions or pageviews) I arriving one by one. Upon arrival of an online
item i, a set Si of eligible bins (fixed node neighbors) for the item is revealed, together
with a weight wia for eligible bin a ∈ Si. The problem is to assign each item i to an
eligible bin in Si or discard it online, while respecting the capacity constraints, so bin
a gets at most n(a) online items. The goal is to maximize both the cardinality of the
allocation (i.e. the total number of assigned items) and the sum of the weights of the
allocated online items.
It was shown in [35] that achieving any positive approximation guarantee for the
total weight of the allocation requires the free disposal assumption, i.e. that there is no
penalty for assigning more online nodes to a bin than its capacity, though these extra
nodes do not count towards the objective. In the advertising application, this means
that in the presence of a contract for n(a) impressions, advertisers are only pleased by
42
– or at least indifferent to – getting more than n(a) impressions. More specifically, if a
set Ia of online items are assigned to each bin a, and Ia(k) denotes the set of k online
nodes with maximum weight in Ia, the goal is to simultaneously maximize cardinality
which is∑
a∈A min(|Ia|, n(a)), and total weight which is∑
a∈A∑
i∈Ia(n(a))wia.
Throughout this chapter, we use Wopt to denote the maximum weight matching,
and overload this notation to also refer to the weight of this matching. Similarly, we
use Copt to denote both the maximum cardinality matching and its cardinality. Note
that Copt and Wopt may be distinct matchings. We aim to find (α, β)-approximations
for the bicriteria online matching problem: These are matchings with weight at least
αWopt and cardinality at least βCopt. Our approach is to study parametrized approx-
imation algorithms that allow a smooth tradeoff curve between the two objectives,
and prove both approximation and hardness results in this framework. As an offline
problem, the above bicriteria problem can be solved optimally in polynomial time,
i.e., one can check if there exists an assignment of cardinality c and weight w respect-
ing capacity constraints. (One can verify this by observing that the integer linear
programming formulation for the offline problem is totally unimodular, and therefore
the problem can be solved by solving the corresponding LP relaxation.) However in
the online competitive setting, even maximizing one of these two objectives does not
admit better than a 1 − 1/e approximation [56]. A naive greedy algorithm gives a
12-approximation for maximizing a single objective, either for cardinality or for total
weight under the free disposal assumption.
3.1 Results and Techniques
The seminal result of Karp, Vazirani and Vazirani [56] gives a simple randomized (1−
1/e)-competitive algorithm for maximizing cardinality. For the weight objective, no
algorithm better than the greedy 1/2-approximation is known, but for the case of large
capacities, a 1−1/e-approximation has been developed [35] following the primal-dual
analysis framework of Buchbinder et al. [18, 65]. Using these results, one can easily
get a (p2, (1−p)(1− 1
e))-approximation for the bicriteria online matching problem with
43
small capacities, and a(p(1− 1
e), (1− p)(1− 1
e))-approximation for large capacities.
These factors are achieved by applying the online algorithm for weight, WeightAlg, and
the online algorithm for cardinality, CardinalityAlg, as subroutines as follows: When
an online item arrives, pass it to WeightAlg with probability p, and CardinalityAlg with
probability 1− p. As for a hardness result, it is easy to show that an approximation
factor better than (α, 1 − α) is not achievable for any α > 0. There is a large gap
between the above approximation factors and hardness results. For example, the
naive algorithm gives a (0.4(1− 1/e), 0.6(1− 1/e)) ≈ (0.25, 0.38)-approximation, but
the hardness result does not preclude a (0.4, 0.6)-approximation. In this chapter, we
tighten the gap between these lower and upper bounds, and present new tradeoff
curves for both algorithms and hardness results. Our lower and upper bound results
are summarized in Figure 3-1. For the case of large capacities, these upper and lower
bound curves are always close (with a maximum vertical gap of 9%), and exactly
coincide at the point (0.43, 0.43).
We first describe our hardness results. In fact, we prove three separate inapprox-
imability results which can be combined to yield a ‘hardness curve’ for the problem.
The first result gives better upper bounds for large values of β; this is based on
structural properties of matchings, proving some invariants for any online algorithm
on a family of instances, and writing a factor-revealing mathematical program (see
Section 3.2.1). The second main result is an improved upper bound for large values
of α, and is based on a new family of instances for which achieving a large value for
α implies very small values of β (see Section 3.2.2). Finally, we show that for any
achievable (α, β), we have α + β ≤ 1− 1e2
(see Theorem 3.2.2).
These hardness results show the limit of what can be achieved in this model. We
next turn to algorithms, to see how close we can come to these limits. The key to our
new algorithmic results lies in the fact that though each subroutine WeightAlg and
CardinalityAlg only receives a fraction of the online items, it can use the entire set of
bins. This may result in both subroutines filling up a bin, but if WeightAlg places t
items in a bin, we can discard t of the items placed there by CardinalityAlg and still
get at least the cardinality obtained by CardinalityAlg and the weight obtained by
44
Figure 3-1: New curves for upper and lower bounds.
WeightAlg. Each subroutine therefore has access to the entire bin capacity, which is
more than it ‘needs’ for those items passed to it. Thus, its competitive ratio can be
made better than 1 − 1/e. For large capacities, we prove the following theorem by
extending the primal-dual analysis of Buchbinder et al. and Feldman et al. [35, 65,
18].
Theorem 3.1.1. For all 0 < p < 1, there is an algorithm for the bicriteria online
matching problem with competitive ratios tending to(p(1− 1
e1/p), (1− p)(1− 1
e1/(1−p)))
as minan(a) tends to infinity.
For small capacities, our result is more technical and is based on studying struc-
tural properties of matchings, proving invariants for our online algorithm over any
instance, and solving a factor-revealing LP that combines these new invariants and
previously known combinatorial techniques by Karp, Vazirani, Vazirani, and Birn-
baum and Mathieu [56, 17]. Factor revealing LPs have been used in the context of
online allocation problems [65, 64]. In our setting, we need to prove new variants
and introduce new inequalities to take into account and analyze the tradeoff between
the two objective functions. This result can also be parametrized by p, the fraction
45
of items sent to WeightAlg, but we do not have a closed form expression. Hence, we
state the result for p = 1/2.
Theorem 3.1.2. For all 0 ≤ p ≤ 1, the approximation guarantee of our algorithm
for the bicriteria online matching problem is lower bounded by the green curve of
Figure 3-1. In particular, for p = 1/2, we have the point (1/3, 0.3698).
Related Work. Our work is related to online ad allocation problems, including
the Display Ads Allocation (DA) problem [35, 34, 2, 76], and the AdWords (AW)
problem [65, 23]. In both of these problems, the publisher must assign online im-
pressions to an inventory of ads, optimizing efficiency or revenue of the allocation
while respecting pre-specified contracts. The Display Ad (DA) problem is the online
matching problem described above only considering the weight objective [35, 15, 24].
In the AdWords (AW) problem, the publisher allocates impressions resulting from
search queries. Advertiser a has a budget B(a) on the total spend instead of a bound
n(a) on the number of impressions. Assigning impression i to advertiser a consumes
wia units of a’s budget instead of 1 of the n(a) slots, as in the DA problem. For both
of these problems, 1− 1e-approximation algorithms have been designed under the as-
sumption of large capacities [65, 18, 35]. None of the above papers for the adversarial
model study multiple objectives at the same time.
Besides the adversarial model studied in this chapter, online ad allocations have
been studied extensively in various stochastic models. In particular, the problem
has been studied in the random order model, where impressions arrive in a random
order [23, 34, 2, 76, 55, 64, 67]; and the iid model in which impressions arrive iid ac-
cording to a known (or unknown) distribution [36, 66, 46, 25, 26]. In such stochastic
settings, primal and dual techniques have been applied to getting improved approxi-
mation algorithms. These techniques are based on computing offline optimal primal
or dual solutions of an expected instance, and using this solution online [36, 23]. It is
not hard to generalize these techniques to the bicritera online matching problem. In
this chapter, we focus on the adversarial model. Note that in order to deal with traffic
spikes, adversarial competitive analysis is important from a practical perspective, as
46
discussed in [67].
Most previous work on online problems with multiple objectives has been in the
domain of routing and scheduling, and with different models. Typically, goals are to
maximize throughput and fairness; see the work of Goel et al. [43, 42], Buchbinder
and Naor [19], and Wang et al. [78]. In this literature, different objectives often come
from applying different functions on the same set of inputs, such as processing times
or bandwidth allocations. In a model more similar to ours, Bilo et al. [16] consider
scheduling where each job has two different and unrelated requirements, processing
time and memory; the goal is to minimize makespan while also minimizing maximum
memory requirements on each machine. In another problem with distinct metrics,
Flammini and Nicosia [38] consider the k-server problem with a distance metric and
time metric defined on the set of service locations. However, unlike our algorithms,
theirs do not compete simultaneously against the best solution for each objective;
instead, they compete against offline solutions that must simultaneously do well on
both objectives. Further, the competitive ratio depends on the relative values of the
two objectives. Such results are of limited use in advertising applications, for instance,
where click-through rates per impression may vary by several orders of magnitude.
3.2 Hardness Instances
In this section for any 0 ≤ α ≤ 1−1/e, we prove upper bounds on β such that the bi-
criteria online matching problem admits an (α, β)-approximation. Note that it is not
possible to achieve α-approximation guarantee for the total weight of the allocation
for any α > 1 − 1/e. We have two types of techniques to achieve upper bounds: a)
Factor-Revealing Linear Programs, b) Super Exponential Weights Instances, which
are discussed in Subsections 3.2.1, and 3.2.2 respectively. Factor revealing LP hard-
ness instances give us the red upper bound curve in Figure 3-1. The orange upper
bound curve in Figure 3-1 is proved by Super Exponential Weights Instances pre-
sented in Subsection 3.2.2, and the black upper bound line in Figure 3-1 is proved in
Theorem 3.2.2.
47
3.2.1 Better Upper Bounds via Factor-Revealing Linear Pro-
grams
We construct an instance, and a linear program LPα,β based on the instance where
α and β are two parameters in the linear program. We prove that if there exists an
(α, β)-approximation for the bicriteria online matching problem, we can find a feasible
solution for LPα,β based on the algorithm’s allocation for the generated instance.
Finally we find out for which pairs (α, β) the linear program LPα,β is infeasible.
These pairs (α, β) are upper bounds for the bicriteria online matching problem.
For any two integers C, l, and some large weight W 4l2, we construct the
instance as follows. We have l phases, and each phase consists of l sets of C identical
items, i.e. l2C items in total. For any 1 ≤ t, i ≤ l, we define Ot,i to be the set i in
phase t that has C identical items. In each phase, we observe the sets of items in
increasing order of i. There are two types of bins: a) l weight bins b1, b2, · · · , bl which
are shared between different phases, b) l2 cardinality bins b′t,i1≤t,i≤l. For each phase
1 ≤ t ≤ l, we have l separate bins b′t,i1≤i≤l. The capacity of all bins is C. We pick
two permutations πt, σt ∈ Sn uniformly at random at the beginning of each phase t
to construct edges. We note that these permutations are private knowledge, and they
are not revealed to the algorithm. For any 1 ≤ i ≤ j ≤ l, we put an edge between
every item in set Ot,i and bin b′t,σt(j) with weight 1 where σt(j) is the jth number in
permutation σt. We also put an edge between every item in set Ot,i and bin bπt(j) (for
each j ≥ i) with weight W t.
Suppose there exists an (α, β)-approximation algorithm Aα,β for the bicriteria on-
line matching problem. For any 1 ≤ t, i ≤ l, let xt,i be the expected number of items
in set Ot,i that algorithm Aα,β assigns to weight bins bπt(j)lj=i. Similarly we define
yt,i to be the expected number of items in set Ot,i that algorithm Aα,β assigns to car-
dinality bins b′t,σt(j)lj=i. We know that when set Ot,i arrives, although the algorithm
can distinguish between weight and cardinality bins, it sees no difference between
the weight bins bπt(j)lj=i, and no difference between the cardinality bins b′t,σt(j)lj=i.
By uniform selection of π and σ, we ensure that in expectation the xt,i items are
48
allocated equally to weight bins bπt(j)lj=i, and the yt,i items are allocated equally
to cardinality bins b′t,σt(j)lj=i. In other words, for 1 ≤ i ≤ j ≤ l, in expectation
xt,i/(l− i+ 1) and yt,i/(l− i+ 1) items of set Ot,i is allocated to bins bπt(j) and b′t,σt(j),
respectively. It is worth noting that similar ideas have been used in previous papers
on online matching [56, 17].
Since weights of all edges to cardinality bins are 1, we can assume that the items
assigned to cardinality bins are kept until the end of the algorithm, and they will not
be thrown away. We can similarly say that the weights of all items for weight bins
is the same in a single phase, so we can assume that an item that has been assigned
to some weight bin in a phase will not be thrown away at least until the end of the
phase. However, the algorithm might use the free disposal assumption for weight
bins in different phases. We have the following capacity constraints on bins bπt(j) and
b′t,σt(j):
∀1 ≤ t, j ≤ l:∑j
i=1 xt,i/(l − i+ 1) ≤ C &∑j
i=1 yt,i/(l − i+ 1) ≤ C. (3.1)
At any stage of phase t, the total weight assigned by the algorithm cannot be
less than α times the optimal weight allocation up to that stage, or we would not
have weight αWopt if the input stopped at this point. After set Ot,i arrives, the
maximum weight allocation achieves at least total weight CiW t which is achieved by
assigning items in set Ot,i′ to weight bin bπt(i′) for each 1 ≤ i′ ≤ i. On the other
hand, the expected weight in allocation of algorithm Aα,β is at most C(tl+W t−1l) +
W t∑i
i′=1 xt,i′ ≤ W t(C/√W +
∑ii′=1 xt,i′). Therefore we have the following inequality
for any 1 ≤ t, i ≤ l:
i∑i′=1
xt,i′/C ≥ αi− 1/√W. (3.2)
We prove in Lemma 3.2.1 that the linear program LPα,β is feasible if there exists
an algorithm Aα,β by defining pi =∑l
t=1 xt,i/lC, and qi =∑l
t=1 yt,i/lC. Now for any
α, we can find the maximum β for which the LPα,β has some feasible solution for
49
large values of l and W . These factor-revealing linear programs yield the red upper
bound curve in Figure 3-1.
LPα,βC1:
∑ii′=1 pi′ ≥ αi− 1/
√W ∀1 ≤ i ≤ l
C2:∑l
i=1 qi ≥ lβ − 1C3: pi + qi ≤ 1 ∀1 ≤ i ≤ l
C4:∑j
i=1 pi/(l − i+ 1) ≤ 1 ∀1 ≤ j ≤ l
C5:∑j
i=1 qi/(l − i+ 1) ≤ 1 ∀1 ≤ j ≤ l
Lemma 3.2.1. If there exists an (α, β)-approximation algorithm for the bicriteria
online matching problem, there exists a feasible solution for LPα,β as well.
Proof. We claim that the values pi =∑l
t=1 xt,i/lC, and qi =∑l
t=1 yt,i/lC form a
feasible solution for LPα,β. Intuitively, pi and qi are the average values of xt,i/C
and yt,i/C in the l different phases. Since we have inequality 3.2 for each value of t,
their average values also admit the same type of inequlity which is constraint C1 in
LPα,β. To prove that constraint C2 holds, we should look at the total cardinality of
algorithm’s allocation in all phases. The optimal cardinality allocation is to assign
set Ot,i to b′t,σt(i) for all 1 ≤ t, i ≤ l which assigns all items and achieves l2C in
cardinality. But algorithm Aα,β assigns at most lC items to weight bins at the end
(after applying free disposals), and∑
1≤t,i≤l yt,i items to cardinality bins. Since it is a
β-approximation for cardinality, we should have that∑
1≤t,i≤l yt,i+ lC ≥ l2Cβ. Based
on definition of qili=1, this inequality is equivalent to constraint C2. Constraint C3
holds because the expected total number of the items the algorithm assigns from each
set to weight and cardinality bins can not be more than C, the number of items in
the set. Constraints C4 and C5 are derived from inequalities of Equation 3.1.
In addition to computational bounds for infeasibility of certain (α, β) pairs, we
can theoretically prove in Theorem 3.2.2 that for any (α, β) with α + β > 1 − 1/e2,
the LPα,β is infeasible so there exists no (α, β) approximation for the problem. We
note that Theorem 3.2.2 is a simple generalization of the 1− 1/e hardness result for
the classic online matching problem [56, 17].
50
Theorem 3.2.2. For any small ε > 0, and α + β ≥ 1 − 1/e2 + ε, there exists no
(α, β)-approximation algorithm for the bicriteria matching problem.
Proof. We just need to show that LPα,β is infeasible. Given a solution of LPα,β, we
find a feasible solution for LP ′ε defined below by setting ri = pi + qi for any 1 ≤ i ≤ l.
LP ′ε∑l
i=1 ri ≥ (1− 1/e2 + ε/2)lri ≤ 1 ∀1 ≤ i ≤ l∑j
i=1 ri/(l − i+ 1) ≤ 2 ∀1 ≤ j ≤ l
The first inequality in LP ′ε is implied by summing up the constraint C1 for i = l,
and constraint C2 in LPα,β, and also using the fact that α+β ≥ (1−1/e2+ε/2)+ε/2.
We note that the ε/2 difference between the α + β and 1 − 1/e2 + ε/2 takes care of
−1/√W and −1 in the right hand sides of constraints C1 and C2 for large enough
values of l and W . Now we prove that LP ′ε is infeasible for any ε > 0 and large
enough l. Suppose there exists a feasible solution r1, r2, · · · , rn. For any pair 1 ≤
i < j ≤ n, if we have ri < 1 and rj > 0, we update the values of ri and rj to
rnewi = ri + min1− ri, rj, and rnewj = rj −min1− ri, rj. Since we are moving the
same amount from rj to ri (for some i < j), all constraints still hold. If we do this
operation iteratively until there is no pair ri and rj with the above properties, we reach
a solution r′ili=1 of this form: 1, 1, · · · , 1, x, 0, 0, · · · , 0 for some 0 ≤ x ≤ 1. Let t be
the maximum index for which r′t is 1. Using the third inequality for j = l, we have that∑ti=1 1/(l−i+1) ≤ 2 which means that ln (l/(l − t+ 1)) ≤ 2. So t is not greater than
l(1−1/e2), and consequently∑l
i=1 r′i ≤ t+1 ≤ (1−1/e2)l+1 < (1−1/e2+ε/2)l. This
contradiction proves that LP ′ε is infeasible which completes the proof of theorem.
3.2.2 Hardness Results for Large Values of Weight Approx-
imation Factor
The factor-revealing linear program LPα,β gives almost tight bounds for small values
of α. In particular, the gap between the the upper and lower bounds for the cardinality
approximation ratio β is less than 0.025 for α ≤ (1 − 1/e2)/2. But for large values
of α (α > (1 − 1/e2)/2), this approach does not give anything better than the α +
51
β ≤ 1 − 1/e2 bound proved in Theorem 3.2.2 . This leaves a maximum gap of
1/e − 1/e2 ≈ 0.23 between the upper and lower bounds at α = 1 − 1/e. In order to
close the gap at α = 1 − 1/e, we present a different analysis based on a new set of
instances, and reduce the maximum gap between lower and upper bounds from 0.23
to less than 0.09 for all values of α ≥ (1− 1/e2)/2.
The main idea is to construct a hardness instance Iγ for any 1/e ≤ γ < 1, and
prove that for any 0 ≤ p ≤ 1−γ, the pair (1−1/e−f(p), p/(1−γ)) is an upper bound
on (α, β) where f(p) is pe(γ+p)
. In other words, there exists no (α, β)-approximation
algorithm for this problem with both α > 1 − 1/e − f(p) and β > p/(1 − γ). By
enumerating different pairs of γ and p, we find the orange upper bound curve in
Figure 3-1.
For any γ ≥ 1/e, we construct instance Iγ as follows: The instance is identical to
the hardness instance in Subsection 3.2.1, but we change some of the edge weights. To
keep the description short, we only describe the edges with modified weights here. Let
r be b0.5 log1/γ lc. In each phase 1 ≤ t ≤ l, we partition the l sets of items Ot,ili=1
into r groups. The first l(1 − γ) sets are in the first group. From the remaining γl
sets, we put the first (1 − γ) fraction in the second group and so on. Formally, we
put set Ot,i in group 1 ≤ z < r for any i ∈ [l− lγz−1 + 1, l− lγz]. Group r of phase t
contains the last lγr−1 sets of items in phase t. The weight of all edges from sets of
items in group z in phase t is W (t−1)r+z for any 1 ≤ z ≤ r and 1 ≤ t ≤ l.
Given an (α, β)-approximation algorithm Aα,β, we similarly define xt,i and yt,i to
be the expected number of items from set Ot,i assigned to weight and cardinality bins
by algorithm Aα,β respectively. We show in the following lemma that in order to have
a high α, the algorithm should allocate a large fraction of sets of items in each group
to the weight bins.
Lemma 3.2.3. For any phase 1 ≤ t ≤ l, and group 1 ≤ z < r, if the expected number
of items assigned to cardinality bins in group z of phase t is at least plCγz−1 (which
is p times the number of all items in groups z, z + 1, · · · , r of phase t), the weight
approximation ratio cannot be greater than 1− 1/e− f(p) where f(p) is pe(γ+p)
.
52
Proof. We define w = W (t−1)r+z which is the weight of edges from items in group z
of phase t to weight bins. At the end of group z of phase t, we look at the expected
number of items allocated to cardinality bins in this group. If it is more than plCγz−1,
we construct a new instance I ′ by changing the weights of the edges of the next items
as follows. Instance I ′ is identical to our instance Iγ up to the end of group z in
phase t. After this group, everything is identical to Iγ except the edge weights that
we change as follows. For every group r ≥ z′ > z in phase t, instead of setting the
weights of edges between items and bins to W (t−1)r+z′ , we set them to the weights
of edges in group z to weight bins which is w = W (t−1)r+z. After phase t, we set
the weights of all edges from items in phases t + 1, t + 2, · · · , l to both weight and
cardinality bins to zero. Since our instance Iγ, and the new instance I ′ are identical
up to the end of group z of phase t, we know that the algorithm has the same expected
allocation up to that point for both instances.
To simplify notation in the rest of the proof, we rename the items in groups
z, z + 1, · · · , r of phase t, and their associated weight bins. We define l′ to be lγz−1
which is the number of sets of items in groups z, z + 1, · · · , r of phase t. For any
1 ≤ i ≤ l′, we let O′i to be the set of items Ot,l−l′+i, and also let b′′i to be the weight
bin bπt(l−l′+i). In instance I ′, the optimum weight allocation is at least wl′C which can
be achieved by assigning set of items O′i to weight bin b′′i for any 1 ≤ i ≤ l′. We also
know that the weight any algorithm achieves in instance I ′ is from assigning these l′
items to the weight bins, because the total weights achieved before and after these l′
sets of items is negligible by the choice of large W . We define ri to be the fraction
of items in set O′i assigned to weight bins b′′jl′j=i for any 1 ≤ i ≤ l′. We want to
prove that if∑(1−γ)l′
i=1 (1 − ri) ≥ pl′, the weight approximation ratio of the algorithm
is at most 1 − 1/e − f(p). We note that group z is the first 1 − γ fraction of these
l′ sets. We prove that even if the algorithm tries to allocate all items to weight bins
b′′i l′i=1, it cannot saturate more than 1− 1/e fraction of these l′ weight bins in total.
Define r′j to be the fraction of weight bin b′′j which is filled with items O′iji=1 for
any 1 ≤ j ≤ l′. When set of items O′i arrive, the algorithm do not see any difference
between items b′′jl′j=i, and we choose their order to be a random permutation. So in
53
expectation, the algorithm cannot saturate more than 1/(l′− i+ 1) fraction of bin b′′j
with set O′i for any 1 ≤ i ≤ j ≤ l′. Therefore for any 1 ≤ j ≤ l′, the fraction r′j is
at most min1,∑j
i=1 1/(l′− i+ 1). Summing up these upper bounds for all l′ values
of j gives us (1− 1/e)l′ upper bound when l′ goes to infinity. Applying the values of
ri’s we get the following bound on the total filled fraction of weight bins b′′jl′j=1:
l′∑j=1
r′j ≤l′∑j=1
min1,j∑i=1
ri/(l′ − i+ 1) =
l′∑j=1
min1,j∑i=1
1/(l′ − i+ 1)−
l′∑j=1
(min1,
j∑i=1
1/(l′ − i+ 1) −min1,j∑i=1
ri/(l′ − i+ 1)
)=
(1− 1/e)l′ −l′(1−1/e)∑j=1
j∑i=1
(1− ri)/(l′ − i+ 1)−
l′∑j=l′(1−1/e)+1
max
0,
l′(1−1/e)∑i=1
(1− ri)/(l′ − i+ 1)−j∑
i′=l′(1−1/e)+1
ri′/(l′ − i′ + 1)
Now we want to upper bound the above expression by (1 − 1/e − f(p))l′ using
the fact that∑l′(1−γ)
i=1 (1 − ri) is at least pl′. It is not hard to see that the above
expression is maximized when ri′ = 1 for any i′ > l′(1 − γ). We also observe that
for any 1 ≤ i ≤ l′(1 − γ), the fraction 1 − ri which is the lack of contribution of
items in set O′i to the weight bins is distributed evenly between bins bjl′j=i. For
the first l′(1 − 1/e) bins, this lack will be counted for sure, but for other bins after
the threshold l′(1 − 1/e), there is some chance that sets of items O′i′l′
i′=l′(1−1/e)+1
cover up for the lack of previous items, and their lacks will not be counted in the
sum. So the maximum of the above expression is achieved when the lacks are for bins
with higher indices, because this way the lack will be distributed more on the bins
after threshold l′(1−1/e) and less before that. We conclude with the following upper
bound on∑l′
j=1 r′j inequality in which ri is zero for i ∈ [l′(1− γ − p), l′(1− γ)] and is
54
1 for other values of i:
(1−1/e)l′−pl′∑k=1
(k + l′(γ − 1/e))
(k + l′γ)−
l′/e∑k=1
(max
0,
pl′∑k′=1
1
(k′ + l′γ)−
k∑k′′=1
1
(l′/e− k′′ + 1)
)
For large values of l and therefore l′, we can write each sum in the above formula
in an integral form by defining variable x to be the index of each sum divided by l′.
We will have the above formula equal to:
l′(
1− 1/e−∫ p
x=0
x+ γ − 1/e
x+ γdx−
∫ p∗
x=0
(ln(
γ + p
γ)− ln(
1/e
1/e− x)
)dx
)
= l′(
1− 1/e− (p− ln(p+ γ
γ)/e)− p
e(γ + p)ln(
γ + p
γ) +
1
e(1 +
γ ln(γ/((γ + p)e))
γ + p)
)
= 1− 1/e− p
e(γ + p)
where p∗ is some fraction such that for x = p∗ we have that γ+pγ
= 1/e1/e−x . In
other words, we should take the integral until the point that the maximum of zero
and the difference of the two summations becomes zero. Computing the integrals,
and summing them up gives us a simple formula 1− 1/e− p/e(γ + p) for the above
expression.
We conclude this part with the main result of this subsection:
Theorem 3.2.4. For any small ε > 0, 1/e ≤ γ < 1, and 0 ≤ p ≤ 1−γ, any algorithm
for bicriteria online matching problem with weight approximation guarantee, α, at
least 1− 1/e− f(p) cannot have cardinality approximation guarantee, β, greater than
p/(1− γ) + ε.
55
Proof. Using Lemma 3.2.3, for any group 1 ≤ z < r in any phase 1 ≤ t ≤ l,
we know that at most p fraction of items are assigned to cardinality bins, because
1−1/e−f(p) is a strictly increasing function in p. Since in each phase the number of
items is decreasing with a factor of γ in consecutive groups, the total fraction of items
assigned to cardinality bins is at most p+pγ+pγ2+· · ·+pγr−2 plus the fraction of items
assigned to cardinality in the last group r of phase t. Even if the algorithm assigns
all of group r to cardinality, it does not achieve more than fraction γr−1 from these
items in each phase. Since the optimal cardinality algorithm can match all items,
the cardinality approximation guarantee is at most p(1 + γ + γ2 + · · ·+ γr−2) + γr−1.
For large enough l (and consequently large enough r), this sum is not more than
p/(1− γ) + ε.
One way to compute the best values for p and γ corresponding to the best upper
bound curve is to solve some corresponding complex equations explicitly. Instead, we
compute these values numerically by trying different values of p and γ which, in turn,
yield the orange upper bound curve in Figure 3-1.
3.3 Algorithm for Large Capacities
We now turn to algorithms, to see how close one can come to matching the upper
bounds of the previous section. In this section, we assume that the capacity n(a) of
each bin a ∈ A is “large”, and give an algorithm with the guarantees in Theorem 3.1.1
as mina∈A n(a)→∞.
Recall that our algorithm Alg uses two subroutines WeightAlg and CardinalityAlg,
each of which, if given an online item, suggests a bin to place it in. Each item
i is independently passed to WeightAlg with probability p and CardinalityAlg with
the remaining probability 1 − p. First note that CardinalityAlg and WeightAlg are
independent and unaware of each other; each of them thinks that the only items
which exist are those passed to it. This allows us to analyze the two subroutines
separately.
We now describe how Alg uses the subroutines. If WeightAlg suggests matching
56
item i to a bin a, we match i to a. If a already has n(a) items assigned to it in total,
we remove any item assigned by CardinalityAlg arbitrarily; if all n(a) were assigned by
WeightAlg, we remove the item of lowest value for a. If CardinalityAlg suggests match-
ing item i to a′, we make this match unless a′ has already had at least n(a′) total items
assigned to it by both subroutines. In other words, the assignments of CardinalityAlg
might be thrown away by some assignments of WeightAlg; however, the total number
of items in a bin is always at least the the number assigned by CardinalityAlg. Items
assigned by WeightAlg are never thrown away due to CardinalityAlg; they may only
be replaced by later assignments of WeightAlg. Thus, we have proved the following
proposition.
Proposition 3.3.1. The weight and cardinality of the allocation of Alg are respec-
tively at least as large as the weight of the allocation of WeightAlg and the cardinality
of the allocation of CardinalityAlg.
Note that the above proposition does not hold for any two arbitrary weight func-
tions, and this is where we need one of the objectives to be cardinality. We now
describe WeightAlg and CardinalityAlg, and prove Theorem 3.1.1. WeightAlg is essen-
tially the exponentially-weighted primal-dual algorithm from [35], which was shown
to achieve a 1 − 1e
approximation for the weighted online matching problem with
large degrees. For completeness, we present the primal and dual LP relaxations for
weighted matching below, and then describe the algorithm. In the primal LP, for
each item i and bin a, variable xia denotes whether impression i is one of the n(a)
most valuable items for bin a.
Primal
max∑
i,awiaxia∑a xia ≤ 1 (∀ i)∑i xia ≤ n(a) (∀ a)
xia ≥ 0 (∀ i, a)
Dual
min∑
a n(a)βa +∑
i zi
βa + zi ≥ wia (∀i, a)
βa, zi ≥ 0 (∀i, a)
Following the techniques of Buchbinder et al. [18], the algorithm of [35] simul-
taneously maintains feasible solutions to both the primal and dual LPs. Each dual
57
variable βa is initialized to 0. When item i arrives online:
• Assign i to the bin a′ = arg maxawia− βa. (If this quantity is negative for all
a, discard i.)
• Set xia′ = 1. If a′ previously had n(a′) items assigned to it, set xi′a′ = 0 for the
least valuable item i′ previously assigned to a′.
• In the dual solution, set zi = wia′−βa′ and update dual variable βa′ as described
below.
Definition 3.3.2 (Exponential Weighting). Let w1, w2, . . . wn(a) be the weights of the
n(a) items currently assigned to bin a, sorted in non-increasing order, and padded
with 0s if necessary.
Set βa = 1
p·n(a)·((1+1/p·n(a))n(a)−1)
∑n(a)j=1 wj
(1 + 1
p·n(a)
)j−1.
Lemma 3.3.3. If WeightAlg is the primal-dual algorithm, with dual variables βa
updated acccording to the exponential weighting rule defined above, the total weight
of the allocation of WeightAlg is at least p ·(1− 1
k
)where k =
(1 + 1
p·d
)d, and d =
minan(a). Note that limd→∞ k = e1/p.
Before proving Lemma 3.3.3, we provide some brief intuition. If all items are
passed to WeightAlg, it was proved in [35] that the algorithm has competitive ratio
tending to 1−1/e as d = minan(a) tends to∞; this is the statement of Lemma 3.3.3
when p = 1. Now, suppose each item is passed to WeightAlg with probability p. The
expected value of the optimum matching induced by those items passed to WeightAlg
is at least p ·Wopt, and this is nearly true (up to o(1) terms) even if we reduce the
capacity of each bin a to p ·n(a). This follows since Wopt assigns at most n(a) items to
bin a, and as we are unlikely to sample more than p·n(a) of these items for the reduced
instance, we do not lose much by reducing capacities. But note that WeightAlg can
use the entire capacity n(a), while there is a solution of value close to pWopt even
with capacities p · n(a). This extra capacity allows an improved competitive ratio of
1− 1e1/p
, proving the lemma.
58
Proof of Lemma 3.3.3. We construct a feasible dual solution as follows. Recall that
the dual LP has a variable βa for each bin a and zi for each online item i. Though
WeightAlg is unaware of items which are not passed to it, we maintain dual variables
for these items as well, purely for the purpose of analysis. Let S denote the set of
items passed to WeightAlg, and I − S those passed to CardinalityAlg. For ease of
notation, we use e1/p to represent(
1 + 1p·d
)dwhere d = minan(a); as d tends to
infinity, the latter expression tends to e1/p.
For each item i, whether it is in S or not, we set zi = maxawia−βa, or zi = 0 if
wia − βa is negative for each bin a. For those items i ∈ S, if zi is positive, we update
βa using the update rule of Definition 3.3.2. This gives a feasible solution to the dual
LP defined on the entire set of items (including those in I − S).
In the previous analysis of weighted online matching in [35], following previous
work of Buchbinder et al. [18], one shows a competitive ratio of 1 − 1/e by arguing
that the change in the primal is at least (1 − 1/e) times the change in the dual at
each step. Since we end with a feasible primal and dual, the total value obtained by
the algorithm (the value of the primal solution) is at least (1− 1/e) ·Wopt.
Here, however, we do not compare the change in the primal to the change in
the dual directly. This is because, when an item in I − S arrives, there is a change
in the dual, but perhaps no change in the primal as the item does not get passed
to WeightAlg. To deal with this, we introduce a new ‘reduced’ objective function∑a∈A p · n(a)βa +
∑i∈S zi. We show that the change in primal is comparable to the
change in this new reduced objective. Though new reduced objective is not an upper
bound on the true optimal solution, it is not too difficult to see that it is at least
p ·Wopt: For the terms involving variables βa, we have simply scaled by p, but the
reduced objective only includes the terms zi for i ∈ S; this latter difference in the
objectives requires some care.
First, though, we examine the relationship between the primal and the reduced
objective. Consider how these change when an item i is sent to WeightAlg. If i is
unassigned (if each wia − βa is non-positive), we have zi = 0 and no change in either
primal or dual. Otherwise, let a be the bin that item i is assigned to, and let v be
59
the value of the currently lowest-valued item assigned to bin a. The change in the
primal is wia − v.
The change in the reduced objective is zi + p · n(a) times the change in βa. Let
βn, βo represent the new and old values of βa respectively. We assume that i now
becomes the most valuable item for bin a.1 Then, from the definition of our update
rule for βa, we have:
βn =
(1 +
1
p · n(a)
)βo +
wia(p · n(a))(e1/p − 1)
− v · e1/p
(p · n(a))(e1/p − 1)
Overall, then, the change in the reduced objective is:
zi + p · n(a) (βn − βo) = (wia − βo) +
(βo +
wiae1/p − 1
− v · e1/p
e1/p − 1
)= wia +
wiae1/p − 1
− v · e1/p
e1/p − 1
= (wia − v) ·(
e1/p
e1/p − 1
)= (wia − v) /
(1− 1
e1/p
)
Since the change in the primal is wia − v, we have that the change in the primal
is at least 1 − 1/e1/p times the change in the reduced objective. It remains only to
argue that the reduced objective is, in expectation, at least p times the original dual
objective. Recall that the reduced objective is∑
a∈A p · n(a)βa +∑
i∈S zi, while the
original dual objective is∑
a∈A n(a)βa +∑
i∈I zi.
The terms involving βa are simply scaled by p in the reduced objective, so for
every run of the algorithm, the contribution of these terms to the reduced objective is
exactly p times that to the original dual objective. However, of the terms involving zis,
the reduced objective only includes a subset which depends on which items are selected
for S. As each item is selected for S with probability p independently, we would like to
conclude that in expectation (though clearly not in every run),∑
i∈S zi = p∑
i∈I zi.
1It is easy to observe that this is the worst case for our algorithm, as this gives the maximumincrease in the dual; see [35].
60
However, the value of zi depends on the random allocation of items to S, since it
depends on each βa at the time item i arrives, and each βa is affected only by those
items in S. Still, it is possible to use linearity of expectation since zi is not a function
of whether i itself is assigned to S (in all cases, zi = wia−βa), only a function of which
previous items were assigned to S. Thus, the expected contribution of each item to
the reduced objective is exactly p times its expected contribution to the original dual
objective; this can be verified by some straightforward algebra which we omit.
Algorithm CardinalityAlg is identical to WeightAlg, except that it assumes all items
have weight 1 for each bin. Since items are assigned to CardinalityAlg with probability
1 − p, Lemma 3.3.3 implies the following corollary. This concludes the proof of
Theorem 3.1.1.
Corollary 3.3.4. The total cardinality of the allocation of CardinalityAlg is at least
(1−p)·(1− 1
k
), where k =
(1 + 1
(1−p)·d
)d, and d = minan(a). Note that limd→∞ k =
e1/(1−p).
3.4 Algorithm for Small Capacities
We now consider algorithms for the case when the capacities of bins are not large.
Without loss of generality, we assume that the capacity of each bin is one, because
we can think about a bin with capacity c as c identical bins with capacity one. So we
have a set A of bins each with capacity one, and a set of items I arriving online. As
before, we use two subroutines WeightAlg and CardinalityAlg, but the algorithms are
slightly different from those in the previous section.
In WeightAlg, we match item i (that has been passed to WeightAlg) to the bin that
maximizes its marginal value. Formally we match i to bin a = arg maxa∈A(wi,a−wi′,a)
where i′ is the last item assigned to a before item i.
In CardinalityAlg, we run the RANKING algorithm presented in [56]. So
CardinalityAlg chooses a permutation π uniformly at random on the set of bins A,
assigns an item i (that has been passed to it) to the bin a that is available, has the
minimum rank in π, and there is also an edge between i and a.
61
3.4.1 Lower Bounding the Weight Approximation Ratio
Let n = |I| be the number of items. We denote the ith arrived item by i. Let ai
be the bin that i is matched to in Wopt for any 1 ≤ i ≤ n. One can assume that
all unmatched items in the optimum weight allocation are matched with zero-weight
edges to an imaginary bin. So Wopt is equal to∑n
i=1wi,ai . Let S be the set of items
that have been passed to WeightAlg. If WeightAlg matches item i to bin aj for some
j > i, we call this a forwarding allocation (edge) because item j (the match of aj in
Wopt) has not arrived yet. We call it a selected forwarding edge if j ∈ S. We define
the marginal value of assigning item i to bin a to be wia minus the value of any item
previously assigned to a.
Lemma 3.4.1. The weight of the allocation of WeightAlg is at least (p/(p+ 1))Wopt.
Proof. Each forwarding edge will be a selected forwarding edge with probability p
because Pr[j ∈ S] is p for any j ∈ I. Let F be the total weight of forwarding edges
of WeightAlg, where by weight of a forwarding edge, we mean its marginal value (not
the actual weight of the edge). Similarly, we define Fs to be the sum of marginal
values of selected forwarding edges. We have the simple equality that the expected
value of F , E(F ), is E(Fs)/p. We define W ′ and Ws to be the total marginal values
of allocation of WeightAlg, and the sum∑
i∈S wi,ai . We know that E(Ws) is pWopt
because Pr[i ∈ S] is p. We prove that W ′ is at least Ws − Fs.
For every item i that has been selected to be matched by WeightAlg, we get at
least marginal value wi,ai minus the sum of all marginal values of items that have been
assigned to bin a by WeightAlg up to now. If we sum up all these lower bounds on our
gains for all selected items, we get Ws(=∑
i∈S wi,ai) minus the sum of all marginal
values of items that has been assigned to ai before item i arrives for all i ∈ S. The
latter part is exactly the definition of Fs. Therefore W ′ is at least Ws − Fs. We
also know that W ′ ≥ F . Using E[F ] ≥ E[Fs]/p, we have that E(W ′) is at least
E(Ws)− pE(W ′), and this yields the p/(p+ 1) approximation factor.
Corollary 3.4.2. The weight and cardinality approximation guarantees of Alg are
at least p/(p+ 1) and (1− p)/(1− p+ 1) respectively.
62
3.4.2 Factor Revealing Linear Program for CardinalityAlg
Our goal in this subsection is to prove an approximation factor for CardinalityAlg better
than the (1−p)/(1−p+1) bound of Theorem 3.1.2 by formulating a factor-revealing
LP that lower bounds it.
Proof of Theorem 3.1.2. We prove that the cardinality approximation ratio of
CardinalityAlg is lower bounded by the solution of the linear program LPk shown
below for any positive integer k. We state the proof and the linear program LPk for
the simpler case of p = 1/2, and show the necessary changes for general p ∈ [0, 1]
at the end of the proof. Before showing how this LP lower bounds the cardinality
approximation factor, we note that the first three lines of constraints in LPk hold for
the weighted version and in fact give us the 1/3 lower bound. The last inequality is
specific to cardinality and makes an improvement on the lower bound from 1/3 to
almost 0.37.
Minimize: β∀1 < i ≤ k: si ≥ si−1 & sfi ≥ sfi−1 & sbi ≥ sbi−1∀1 ≤ i ≤ k: ti ≥ ti−1 & ti ≥ sfi & si = sfi + sbi
β ≥ sk + tk & β ≥ 1/2− sfk∀1 < i ≤ k: si − si−1 ≥ 1/2k − (si + ti)/k
In CardinalityAlg, a uniformly random permutation π is selected on set A of bins.
Let A′ ⊆ A be the set of bins matched in Copt. We define A′′ to be the set of bins
matched in CardinalityAlg which depends on permutation π. We divide permutation
π into k equal parts each with |A|/k bins. For each a ∈ A′, we define i(a) to be the
match of a in Copt. Let Ac be the set of bins like a such that i(a) has been selected to
be passed to CardinalityAlg. For each 1 ≤ i ≤ k, we define sets Si, and Ti as follows:
Si = A′∩A′′∩Ac∩πj|1 ≤ j ≤ (i|A|/k), and Ti = A′∩A′′∩πj|1 ≤ j ≤ (i|A|/k)\Ac.
In other words, Si is the set of bins in the first i parts of π that are matched in
both Copt and our algorithm, and their matches in Copt are selected to be passed to
CardinalityAlg. The only difference for Ti is that their matches in Copt are not selected
for CardinalityAlg. We also partition Si into two sets: SFi the set of bins that has
been matched in our algorithm with forwarding edges, and the rest in set SBi. We
63
remind that a forwarding edge in our algorithm is an edge matching an item to a bin
a such that i(a) has not arrived yet. We prove that this is a feasible solution for LPk:
si = E[|Si|/|A′|], sfi = E[|SFi|/|A′|], sbi = E[|SBi|/|A′|], and ti = E[|Ti|/|A′|].
Now we prove the constraints of LPk hold. The first 4 constraints are the mono-
tonicity constraints which hold by definition, the same is true for the 6th constraint.
The constraint ti ≥ sfi holds because every forwarding edge (incident to a bin in the
first i parts of π) with probability 1/2 will be counted in sfi, and with the other 1/2
will be counted in ti. This way everything in sfi will be counted, but there might
be other uncounted bins in ti. This is why we get an inequality instead of equality.
We have β at least 1/2 − sfk because in expectation half of the items matched in
Copt = |A′| will be selected for cardinality, and the number of them that are not
matched in CardinalityAlg is at most sfk|A′|. The inequality β ≥ sk + tk holds by
definition. These inequalities give us the 1/3 lower bound on β. We add the follow-
ing inequality as well to get a better approximation ratio for cardinality. We note
that similar (and simpler) inequalities have been presented in the literature of online
matching, e.g. Lemma 5 in [17].
∀1 ≤ i ≤ k, si − si−1 ≥ 1/2k − (si + ti)/k
Let s0 = 0 to make the math consistent. This inequality lower bounds the expected
number of matched bins in A′ ∩A′′ ∩Ac in part i of permutation π. For each a ∈ A′
in part i of permutation π, we prove that the probability that a ∈ Ac, and a is in A′′
is at least 1/2 − (si + ti)/k. With probability 1/2, a is in set Ac. In this case, we
know that if no item is matched to a by CardinalityAlg, item i(a) has been assigned by
CardinalityAlg to some bin a′ whose rank in π is smaller than a. This means that i(a)
is matched to one of the bins in Si or Ti. Since π is selected uniformly at random,
a is selected uniformly at random from A′ ∩ Ac. So the probability of this event
(i(a) being matched to some bin before a in π) given that a ∈ A′ ∩ Ac is at most
(si+ ti)/|A′∩Ac|. Therefore with probability at least (1− (si+ ti)/|A′∩Ac|)/2, bin a
is matched by CardinalityAlg. Summing up these lower bounds on probabilities for all
64
different choices of a, and applying the fact that in expectation there are |A′|/k bins
like a ∈ A′ in part i of π, we get the lower bound on si−si−1 ≥ 1/2k−(si+ti)/2k. We
note that E(|A′∩Ac|) = E(|A′|)/2, and for large values of |A′|, the ratio |A′|/|A′∩Ac|
is around 2 with high precision, i.e. it is at most 2+ε for any small constant ε > 0, and
large enough |A′|. Since the solution for this linear program is greater than 0.3698 for
large enough k, we know that the approximation ratio of CardinalityAlg and therefore
our algorithm is at least 0.3698.
For any other value of p, we can change the LPk as follows. We just need to
change three constraints in LPk. Instead of ti ≥ sfi, we have ti/p ≥ sfi/(1 − p)
because with probability p we choose each item for weight and with the remaining
probability 1 − p for cardinality. With a similar reason, we have β ≥ (1 − p) − sfk,
and si−si−1 ≥ (1−p)/k−(si+ti)/k instead of their‘ simpler versions. We enumerate
on different values of p, and solve the LPk for each of these values to get the green
lower bound curve in Figure 3-1.
65
Chapter 4
Submodular Secretary Problem
and its Extensions
Online auction is the essence of many modern markets, particularly networked mar-
kets, in which information about goods, agents, and outcomes is revealed over a period
of time, and the agents must make irrevocable decisions without knowing future in-
formation. Optimal stopping theory is a powerful tool for analyzing such scenarios
which generally require optimizing an objective function over the space of stopping
rules for an allocation process under uncertainty. Combining optimal stopping the-
ory with game theory allows us to model the actions of rational agents applying
competing stopping rules in an online market. This first has been done by Hajiaghayi
et al. [48] who considered the well-known secretary problem in online settings and
initiated several follow-up papers (see e.g. [7, 8, 9, 47, 52, 59]).
Perhaps the most classic problem of stopping theory is the secretary problem.
Imagine that you manage a company, and you want to hire a secretary from a pool of
n applicants. You are very keen on hiring only the best and brightest. Unfortunately,
you cannot tell how good a secretary is until you interview her, and you must make
an irrevocable decision whether or not to make an offer at the time of the interview.
The problem is to design a strategy which maximizes the probability of hiring the
most qualified secretary. It is well-known since 1963 [27] that the optimal policy is
to interview the first t − 1 applicants, then hire the next one whose quality exceeds
66
that of the first t − 1 applicants, where t is defined by∑n
j=t+11j−1 ≤ 1 <
∑nj=t
1j−1 ;
as n → ∞, the probability of hiring the best applicant approaches 1/e, as does
the ratio t/n. Note that a solution to the secretary problem immediately yields an
algorithm for a slightly different objective function optimizing the expected value of
the chosen element. Subsequent papers have extended the problem by varying the
objective function, varying the information available to the decision-maker, and so
on, see e.g., [3, 41, 75, 80].
An important generalization of the secretary problem with several applications
(see e.g., a survey by Babaioff et al. [8]) is called the multiple-choice secretary problem
in which the interviewer is allowed to hire up to k ≥ 1 applicants in order to maximize
performance of the secretarial group based on their overlapping skills (or the joint
utility of selected items in a more general setting). More formally, assuming applicants
of a set S = a1, a2, · · · , an (applicant pool) arriving in a uniformly random order,
the goal is to select a set of at most k applicants in order to maximize a profit function
f : 2S 7→ R. We assume f is non-negative throughout this chapter. For example,
when f(T ) is the maximum individual value [39, 40], or when f(T ) is the sum of
the individual values in T [59], the problem has been considered thoroughly in the
literature. Indeed, both of these cases are special monotone non-negative submodular
functions that we consider in this chapter. A function f : 2S 7→ R is called submodular
if and only if ∀A,B ⊆ S : f(A) + f(B) ≥ f(A ∪ B) + f(A ∩ B). An equivalent
characterization is that the marginal profit of each item should be non-increasing,
i.e., f(A ∪ a) − f(A) ≤ f(B ∪ a) − f(B) if B ⊆ A ⊆ S and a ∈ S \ B. A
function f : 2S 7→ R is monotone if and only if f(A) ≤ f(B) for A ⊆ B ⊆ S; it is
non-monotone if it is not necessarily the case. Since the number of sets is exponential,
we assume a value oracle access to the submodular function; i.e., for a given set T , an
algorithm can query an oracle to find its value f(T ). As we discuss below, maximizing
a (monotone or non-monotone) submodular function which demonstrates economy of
scale is a central and very general problem in combinatorial optimization and has
been subject of a thorough study in the literature.
The closest setting to our submodular multiple-choice secretary problem is the ma-
67
troid secretary problem considered by Babaioff et al. [9]. In this problem, we are given
a matroid by a ground set U of elements and a collection of independent (feasible)
subsets I ⊆ 2U describing the sets of elements which can be simultaneously accepted.
We recall that a matroid has three properties: 1) the empty set is independent; 2)
every subset of an independent set is independent (closed under containment)1; and
finally 3) if A and B are two independent sets and A has more elements than B, then
there exists an element in A which is not in B and when added to B still gives an
independent set2. The goal is to design online algorithms in which the structure of U
and I is known at the outset (assume we have an oracle to answer whether a subset
of U belongs to I or not), while the elements and their values are revealed one at
a time in random order. As each element is presented, the algorithm must make an
irrevocable decision to select or reject it such that the set of selected elements belongs
to I at all times. Babaioff et al. present an O(log r)-competitive algorithm for general
matroids, where r is the rank of the matroid (the size of the maximal independent
set), and constant-competitive algorithms for several special cases arising in practical
scenarios including graphic matroids, truncated partition matroids, and bounded de-
gree transversal matroids. However, they leave as a main open question the existence
of constant-competitive algorithms for general matroids. Our constant-competitive
algorithms for the submodular secretary problem in this chapter can be considered
in parallel with this open question. To generalize both results of Babaioff et al. and
ours, we also consider the submodular matroid secretary problem in which we want
to maximize a submodular function over all independent (feasible) subsets I of the
given matroid. Moreover, we extend our approach to the case in which l matroids
are given and the goal is to find the set of maximum value which is independent with
respect to all the given matroids. We present an O(l log2 r)-competitive algorithm for
the submodular matroid secretary problem generalizing previous results.
Prior to our work, there was no polynomial-time algorithm with a nontrivial guar-
antee for the case of l matroids—even in the offline setting—when l is not a fixed
1This is sometimes called the hereditary property.2This is sometimes called the augmentation property or the independent set exchange property.
68
constant. Lee et al. [61] give a local-search procedure for the offline setting that runs
in time O(nl) and achieves approximation ratio l+ε. Even the simpler case of having a
linear function cannot be approximated to within a factor better than Ω(l/ log l) [51].
Our results imply an algorithm with guarantees O(l log r) and O(l log2 r) for the of-
fline and (online) secretary settings, respectively. Both these algorithms run in time
polynomial in l. In case of the knapsack constraints, the only previous relevant work
that we are aware of is that of Lee et al. [61] which gives a (5+ε) approximation in the
offline setting if the number of constraints is a constant. In contrast, our results work
for arbitrary number of knapsack constraints, albeit with a loss in the guarantee; see
Theorem 4.1.3.
Our competitive ratio for the submodular secretary problem is 71−1/e . Though
our algorithm is relatively simple, it has several phases and its analysis is relatively
involved. As we point out below, we cannot obtain any approximation factor better
than 1 − 1/e even for offline special cases of our setting unless P = NP. A natural
generalization of a submodular function while still preserving economy of scale is a
subadditive function f : 2S 7→ R in which ∀A,B ⊆ S : f(A) + f(B) ≥ f(A ∪ B). In
this chapter, we show that if we consider the subadditive secretary problem instead
of the submodular secretary problem, there is no algorithm with competitive ratio
o(√n). We complement this result by giving an O(
√n)-competitive algorithm for the
subadditive secretary problem.
Background on submodular maximization Submodularity, a discrete analog
of convexity, has played a central role in combinatorial optimization [62]. It appears
in many important settings including cuts in graphs [53, 44, 71], plant location prob-
lems [22, 21], rank function of matroids [28], and set covering problems [30].
The problem of maximizing a submodular function is of essential importance,
with special cases including Max Cut [44], Max Directed Cut [49], hypergraph cut
problems, maximum facility location [1, 22, 21], and certain restricted satisfiability
problems [50, 29]. While the Min Cut problem in graphs is a classical polynomial-
time solvable problem, and more generally it has been shown that any submodular
69
function can be minimized in polynomial time [53, 72], maximization turns out to be
more difficult and indeed all the aforementioned special cases are NP-hard.
Max-k-Cover, where the goal is to choose k sets whose union is as large as possible,
is another related problem. It is shown that a greedy algorithm provides a (1− 1/e)
approximation for Max-k-Cover [58] and this is optimal unless P = NP [30]. More
generally, we can view this problem as maximization of a monotone submodular func-
tion under a cardinality constraint, that is, we seek a set S of size k maximizing f(S).
The greedy algorithm again provides a (1− 1/e) approximation for this problem [69].
A 1/2 approximation has been developed for maximizing monotone submodular func-
tions under a matroid constraint [37]. A (1 − 1/e) approximation has been also ob-
tained for a knapsack constraint [73], and for a special class of submodular functions
under a matroid constraint [20].
Recently constant factor (34
+ ε)-approximation algorithms for maximizing non-
negative non-monotone submodular functions has also been obtained [32]. Typical
examples of such a problem are max cut and max directed cut. Here, the best
approximation factors are 0.878 for max cut [44] and 0.859 for max directed cut [29].
The approximation factor for max cut has been proved optimal, assuming the Unique
Games Conjecture [57]. Generalizing these results, Vondrak very recently obtains a
constant factor approximation algorithm for maximizing non-monotone submodular
functions under a matroid constraint [77]. Subadditive maximization has been also
considered recently (e.g. in the context of maximizing welfare [31]).
Submodular maximization also plays a role in maximizing the difference of a
monotone submodular function and a modular function. A typical example of this
type is the maximum facility location problem in which we want to open a subset of
facilities and maximize the total profit from clients minus the opening cost of facilities.
Approximation algorithms have been developed for a variant of this problem which
is a special case of maximizing nonnegative submodular functions [1, 22, 21]. The
current best approximation factor known for this problem is 0.828 [1]. Asadpour et
al. [5] study the problem of maximizing a submodular function in a stochastic setting,
and obtain constant-factor approximation algorithms.
70
4.1 Our Results and Techniques
The main theorem in this chapter is as follows.
Theorem 4.1.1. There exists a 71−1/e-competitive algorithm for the monotone sub-
modular secretary problem. More generally there exists a 8e2-competitive algorithm
for the non-monotone submodular secretary problem.
We prove Theorem 4.1.1 in Section 4.2. We first present our simple algorithms
for the problem. Since our algorithm for the general non-monotone case uses that of
monotone case, we first present the analysis for the latter case and then extend it for
the former case. We divide the input stream into equal-sized segments, and show that
restricting the algorithm to pick only one item from each segment decreases the value
of the optimum by at most a constant factor. Then in each segment, we use a standard
secretary algorithm to pick the best item conditioned on our previous choices. We
next prove that these local optimization steps lead to a global near-optimal solution.
The argument breaks for the non-monotone case since the algorithm actually
approximates a set which is larger than the optimal solution. The trick is to invoke
a new structural property of (non-monotone) submodular functions which allows us
to divide the input into two equal portions, and randomly solve the problem on one.
Indeed Theorem 4.1.1 can be extended for the submodular matroid secretary
problem as follows.
Theorem 4.1.2. There exists an O(l log2 r) competitive algorithm for the (non-
monotone) matroid submodular secretary problem, where r is the maximum rank of
the given l matroids.
We prove theorem 4.1.2 in Section 4.3. We note that in the submodular matroid
secretary problem, selecting (bad) elements early in the process might prevent us
from selecting (good) elements later since there are matroid independence (feasibility)
constraints. To overcome this issue, we only work with the first half of the input. This
guarantees that at each point in expectation there is a large portion of the optimal
solution that can be added to our current solution without violating the matroid
71
constraint. However, this set may not have a high value. As a remedy we prove there
is a near-optimal solution all of whose large subsets have a high value. This novel
argument may be of its own interest.
We shortly mention in Section 4.4 our results for maximizing a submodular sec-
retary problem with respect to l knapsack constraints. In this setting, there are l
knapsack capacities Ci : 1 ≤ i ≤ l, and each item j has different weights wij associ-
ated with each knapsack. A set T of items is feasible if and only if for each knapsack
i, we have∑
j∈T wij ≤ Ci.
Theorem 4.1.3. There exists an O(l)-competitive algorithm for the (non-monotone)
multiple knapsack submodular secretary problem, where l denotes the number of given
knapsack constraints.
Lee et al. [61] gives a better (5 + ε) approximation in the offline setting if l is a
fixed constant.
We next show that indeed submodular secretary problems are the most general
cases that we can hope for constant competitiveness.
Theorem 4.1.4. For the subadditive secretary problem, there is no algorithm with
competitive ratio in o(√n). However there is an algorithm with almost tight O(
√n)
competitive ratio in this case.
We prove Theorem 4.1.4 in Section 4.5. The algorithm for the matching upper
bound is very simple, however the lower bound uses clever ideas and indeed works
in a more general setting. We construct a subadditive function, which interestingly
is almost submodular, and has a “hidden good set”. Roughly speaking, the value
of any query to the oracle is proportional to the intersection of the query and the
hidden good set. However, the oracle’s response does not change unless the query has
considerable intersection with the good set which is hidden. Hence, the oracle does
not give much information about the hidden good set.
Finally in our concluding remarks in Section 4.6, we briefly discuss two other
aggregate functions max and min, where the latter is not even submodular and models
a bottle-neck situation in the secretary problem.
72
Remark Subsequent to our study of online submodular maximization [12], Gupta et
al. [45] consider similar problems. By reducing the case of non-monotone submodular
functions to several runs of the greedy algorithm for monotone submodular functions,
they present O(p)-approximation algorithms for maximizing submodular functions (in
the offline setting) subject to p-independence systems (which include the intersection
of p matroids), and constant factor approximation algorithms when the maximization
is subject to a knapsack constraint. In the online secretary setting, they provide O(1)-
competitive results for maximizing a submodular function subject to cardinality or
partition matroid constraints. They also obtain an O(log r) competitive ratio for
maximization subject to a general matroid of rank r. The latter result improves our
Theorem 4.1.2 when l = 1.
4.2 The Submodular Secretary Problem
4.2.1 Algorithms
In this section, we present the algorithms used to prove Theorem 4.1.1. In the classic
secretary problem, the efficiency value of each secretary is known only after she arrives.
In order to marry this with the value oracle model, we say that the oracle answers
the query regarding the efficiency of a set S ′ ⊆ S only if all the secretaries in S ′ have
already arrived and been interviewed.
Our algorithm for the monotone submodular case is relatively simple though its
analysis is relatively involved. First we assume that n is a multiple of k, since other-
wise we could virtually insert n−kbnkc dummy secretaries in the input: for any subset
A of dummy secretaries and a set B ⊆ S, we have that f(A ∪ B) = f(B). In other
words, there is no profit in employing the dummy secretaries. To be more precise,
we simulate the augmented input in such a way that these secretaries are arriving
uniformly at random similarly to the real ones. Thus, we say that n is a multiple of
k without loss of generality.
We partition the input stream into k equally-sized segments, and, roughly speak-
73
Algorithm 1 Monotone Submodular Secretary Algorithm
Input: A monotone submodular function f : 2S 7→ R, and a randomly permutedstream of secretaries, denoted by (a1, a2, . . . , an), where n is an integer multiple of k.Output: A subset of at most k secretaries.
Let T0 ← ∅Let l← n/kfor i← 1 to k do phase i
Let ui ← (i− 1)l + l/eLet αi ← max
(i−1)l≤j<uif(Ti−1 ∪ aj)
if αi < f(Ti−1) thenαi ← f(Ti−1)
end ifPick an index pi : ui ≤ pi < il such that f(Ti−1 ∪ api) ≥ αiif such an index pi exists then
Let Ti ← Ti−1 ∪ apielse
Let Ti ← Ti−1end if
end forOutput Tk as the solution
ing, try to employ the best secretary in each segment. Let l := nk
denote the
length of each segment. Let a1, a2, · · · , an be the actual ordering in which the
secretaries are interviewed. Break the input into k segments such that Sj =
a(j−1)l+1, a(j−1)l+2, . . . , ajl for 1 ≤ j < k, and Sk = a(k−1)l+1, a(k−1)l+2, . . . , an.
We employ at most one secretary from each segment Si. Note that this way of having
several phases of (almost) equal length for the secretary problem seems novel to this
chapter, since in previous works there are usually only two phases (see e.g. [48]). The
phase i of our algorithm corresponds to the time interval when the secretaries in Si
arrive. Let Ti be the set of secretaries that we have employed from⋃ij=1 Sj. Define
T0 := ∅ for convenience. In phase i, we try to employ a secretary e from Si that maxi-
mizes f(Ti−1∪e)−f(Ti−1). For each e ∈ Si, we define gi(e) = f(Ti−1∪e)−f(Ti−1).
Then, we are trying to employ a secretary x ∈ Si that has the maximum value for
gi(e). Using a classic algorithm for the secretary problem (see [27] for instance) for
employing the single secretary, we can solve this problem with constant probability
1/e. Hence, with constant probability, we pick the secretary that maximizes our local
74
profit in each phase. It leaves us to prove that this local optimization leads to a
reasonable global guarantee.
The previous algorithm fails in the non-monotone case. Observe that the first
if statement is never true for a monotone function, however, for a non-monotone
function this guarantees the values of sets Ti are non-decreasing. Algorithm 2 first
divides the input stream into two equal-sized parts: U1 and U2. Then, with probability
1/2, it calls Algorithm 1 on U1, whereas with the same probability, it skips over the
first half of the input, and runs Algorithm 1 on U2.
Algorithm 2 Submodular Secretary Algorithm
Input: A (possibly non-monotone) submodular function f : 2S 7→ R, and arandomly permuted stream of secretaries, denoted by (a1, a2, . . . , an), where n is aninteger multiple of 2k.Output: A subset of at most k secretaries.
Let U1 := a1, a2, . . . , an/2Let U2 := an/2 + 1, . . . , an−1, anLet 0 ≤ X ≤ 1 be a uniformly random value.if X ≤ 1/2 then
Run Algorithm 1 on U1 to get S1
Output S1 as the solutionelse
Run Algorithm 1 on U2 to get S2
Output S2 as the solutionend if
4.2.2 Analysis
In this section, we prove Theorem 4.1.1. Since the algorithm for the non-monotone
submodular secretary problem uses that for the monotone submodular secretary prob-
lem, first we start with the monotone case.
Monotone Submodular
We prove in this section that for Algorithm 1, the expected value of f(Tk) is within a
constant factor of the optimal solution. Let R = ai1 , ai2 , · · · , aik be the optimal so-
lution. Note that the set i1, i2, · · · , ik is a uniformly random subset of 1, 2, · · · , n
75
with size k. It is also important to note that the permutation of the elements of the
optimal solution on these k places is also uniformly random, and is independent from
the set i1, i2, · · · , ik. For example, any of the k elements of the optimum can appear
as ai1 . These are two key facts used in the analysis.
Before starting the analysis, we present a simple property of submodular functions
which will prove useful in the analysis.
Lemma 4.2.1. If f : 2S 7→ R is a submodular function, we have f(B) − f(A) ≤∑a∈B\A [f(A ∪ a)− f(A)] for any A ⊆ B ⊆ S.
Proof. Let k := |B|− |A|. Then, define in an arbitrary manner sets Biki=0 such that
• B0 = A,
• |Bi \Bi−1| = 1 for i : 1 ≤ i ≤ k,
• and Bk = B.
Let bi := Bi \Bi−1 for i : 1 ≤ i ≤ k. We can write f(B)− f(A) as follows
f(B)− f(A) =k∑i=1
[f(Bi)− f(Bi−1)]
=k∑i=1
[f(Bi−1 ∪ bi)− f(Bi−1)]
≤k∑i=1
[f(A ∪ bi)− f(A)] ,
where the last inequality follows from the non-increasing marginal profit property of
submodular functions. Noticing that bi ∈ B \A and they are distinct, namely bi 6= bi′
for 1 ≤ i 6= i′ ≤ k, finishes the argument.
Define X := Si : |Si ∩ R| 6= ∅. For each Si ∈ X , we pick one element, say si, of
Si ∩ R randomly. These selected items form a set called R′ = s1, s2, · · · , s|X | ⊆ R
of size |X |. Since our algorithm approximates such a set, we study the value of such
random samples of R in the following lemmas. We first show that restricting ourselves
76
to picking at most one element from each segment does not prevent us from picking
many elements from the optimal solution (i.e., R).
Lemma 4.2.2. The expected value of the number of items in R′ is at least k(1−1/e).
Proof. We know that |R′| = |X |, and |X | is equal to k minus the number of sets Si
whose intersection with R is empty. So, we compute the expected number of these
sets, and subtract this quantity from k to obtain the expected value of |X | and thus
|R′|.
Consider a set Sq, 1 ≤ q ≤ k, and the elements of R = ai1 , ai2 , . . . , aik. Define
Ej as the event that aij is not in Sq. We have Pr(E1) = (k−1)ln
= 1 − 1k, and for any
i : 1 < i ≤ k, we get
Pr
(Ei
∣∣∣∣∣i−1⋂j=1
Ej
)=
(k − 1)l − (i− 1)
n− (i− 1)≤ (k − 1)l
n= 1− 1
k,
where the last inequality follows from a simple mathematical fact: x−cy−c ≤
xy
if c ≥ 0
and x ≤ y. Now we conclude that the probability of the event Sq ∩R = ∅ is
Pr(∩ki=1Ei) = Pr(E1) · Pr(E2|E1) · · ·Pr(Ek| ∩k−1j=1 Ej) ≤(
1− 1
k
)k≤ 1
e.
Thus each of the sets S1, S2, . . . , Sk does not intersect with R with probability at
most 1/e. Hence, the expected number of such sets is at most k/e. Therefore, the
expected value of |X | = |R′| is at least k(1− 1/e).
The next lemma materializes the proof of an intuitive statement: if you randomly
sample elements of the set R, you expect to obtain a profit proportional to the size
of your sample. An analog of this is proved in [31] for the case when |R|/|A| is an
integer.
Lemma 4.2.3. For a random subset A of R, the expected value of f(A) is at least
|A|k· f(R).
Proof. Let (x1, x2, . . . , xk) be a random ordering of the elements of R. For r =
1, 2, . . . , k, let Fr be the expectation of f(x1, . . . , xr), and define Dr := Fr − Fr−1,
77
where F0 is interpreted to be equal to zero. Letting a := |A|, note that f(R) = Fk =
D1+· · ·+Dk, and that the expectation of f(A) is equal to Fa = D1+· · ·+Da. We claim
that D1 ≥ D2 ≥ · · · ≥ Dk, from which the lemma follows easily. Let (y1, y2, . . . , yk)
be a cyclic permutation of (x1, x2, . . . , xk), where y1 = xk, y2 = x1, y3 = x2, . . . , yk =
xk−1. Notice that for i < k, Fi is equal to the expectation of f(y2, . . . , yi+1) since
y2, . . . , yi+1 is equal to x1, . . . , xi.
Fi is also equal to the expectation of f(y1, . . . , yi), since the sequence
(y1, . . . , yi) has the same distribution as that of (x1, · · · , xi). Thus, Di+1 is
the expectation of f(y1, . . . , yi+1) − f(y2, . . . , yi+1), whereas Di is the ex-
pectation of f(y1, . . . , yi) − f(y2, . . . , yi). The submodularity of f implies
that f(y1, . . . , yi+1) − f(y2, . . . , yi+1) is less than or equal to f(y1, . . . , yi) −
f(y2, . . . , yi), hence Di+1 ≤ Di.
Here comes the crux of our analysis where we prove that the local optimization
steps (i.e., trying to make the best move in each segment) indeed lead to a globally
approximate solution.
Lemma 4.2.4. The expected value of f(Tk) is at least |R′|
7k· f(R).
Proof. Define m := |R′| for the ease of reference. Recall that R′ is a set of secretaries
s1, s2, . . . , sm such that si ∈ Shi ∩ R for i : 1 ≤ i ≤ m and hi : 1 ≤ hi ≤ k. Also
assume without loss of generality that hi′ ≤ hi for 1 ≤ i′ < i ≤ m, for instance,
s1 is the first element of R′ to appear. Define ∆j for each j : 1 ≤ j ≤ k as the
gain of our algorithm while working on the segment Sj. It is formally defined as
∆j := f(Tj) − f(Tj−1). Note that due to the first if statement in the algorithm,
∆j ≥ 0 and thus E[∆j] ≥ 0. With probability 1/e, we choose the element in Sj
which maximizes the value of f(Tj) (given that the set Tj−1 is fixed). Notice that by
definition of R′ only one si appears in Shi . Since si ∈ Shi is one of the options,
E[∆hi ] ≥E[f(Thi−1 ∪ si)− f(Thi−1)]
e. (4.1)
To prove by contradiction, suppose E[f(Tk)] <m7k· f(R). Since f is monotone,
E[f(Tj)] <m7k· f(R) for any 0 ≤ j ≤ k. Define B := si, si+1, · · · , sm. By
78
Lemma 4.2.1 and monotonicity of f ,
f(B) ≤ f(B ∪ Thi−1) ≤ f(Thi−1) +m∑j=i
[f(Thi−1 ∪ sj)− f(Thi−1)],
which implies
E[f(B)] ≤ E[f(Thi−1)] +m∑j=i
E[f(Thi−1 ∪ sj)− f(Thi−1)].
Since the items in B are distributed uniformly at random, and there is no difference
between si1 and si2 for i ≤ i1, i2 ≤ m, we can say
E[f(B)] ≤ E[f(Thi−1)] + (m− i+ 1) · E[f(Thi−1 ∪ si)− f(Thi−1)]. (4.2)
We conclude from (4.1) and (4.2)
E[∆hi ] ≥E[f(Thi−1 ∪ si)− f(Thi−1)]
e≥ E[f(B)]− E[f(Thi−1)]
e(m− i+ 1).
Since B is a random sample of R, we can apply Lemma 4.2.3 to get E[f(B)] ≥|B|kf(R) = f(R)(m− i+ 1)/k. Since E[f(Thi−1)] ≤ m
7k· f(R), we reach
E[∆hi ] ≥E[f(B)]− E[f(Thi−1)]
e(m− i+ 1)≥ f(R)
ek− m
7kf(R) · 1
e(m− i+ 1). (4.3)
Adding up (4.3) for i : 1 ≤ i ≤ dm/2e, we obtain
dm/2e∑i=1
E[∆hi ] ≥⌈m
2
⌉· f(R)
ek− m
7ek· f(R) ·
dm/2e∑i=1
1
m− i+ 1.
Since∑b
j=a1j≤ ln b
a+1for any integer values of a, b : 1 < a ≤ b, we conclude
dm/2e∑i=1
E[∆hi ] ≥⌈m
2
⌉· f(R)
ek− m
7ek· f(R) · ln m⌊
m2
⌋ .
79
A similar argument for the range 1 ≤ i ≤ bm/2c gives
bm2 c∑i=1
E[∆hi ] ≥⌊m
2
⌋· f(R)
ek− m
7ek· f(R) · ln m⌈
m2
⌉ .We also know that both
∑bm/2ci=1 E[∆hi ] and
∑dm/2ei=1 E[∆hi ] are at most E[f(Tk)]
because f(Tk) ≥∑m
i=1 ∆hi . We conclude with
2E[f(Tk)] ≥⌈m
2
⌉ f(R)
ek− mf(R)
7ek· ln m⌊
m2
⌋ +⌊m
2
⌋ f(R)
ek− mf(R)
7ek· ln m⌈
m2
⌉≥ mf(R)
ek− mf(R)
7ek· ln m2⌊
m2
⌋ ⌈m2
⌉ , and sincem2
bm/2cdm/2e< 4.5
≥ mf(R)
ek− mf(R)
7ek· ln 4.5 =
mf(R)
k·(
1
e− ln 4.5
7e
)≥ mf(R)
k· 2
7,
which contradicts E[f(Tk)] <mf(R)
7k, hence proving the supposition false.
The following theorem wraps up the analysis of the algorithm.
Theorem 4.2.5. The expected value of the output of our algorithm is at least
1−1/e7f(R).
Proof. The expected value of |R′| = m ≥ (1 − 1/e)k from Lemma 4.2.2. In other
words, we have∑k
m=1 Pr[|R′| = m] · m ≥(1− 1
e
)k. We know from Lemma 4.2.4
that if the size of R′ is m, the expected value of f(Tk) is at least m7kf(R), implying
that∑
v∈V Pr[f(Tk) = v
∣∣ |R′| = m]·v ≥ m
7kf(R), where V denotes the set of different
values that f(Tk) can get. We also know that
E[f(Tk)] =k∑
m=1
E[f(Tk)||R′| = m] Pr[|R′| = m] ≥k∑
m=1
m
7kf(R) Pr[|R′| = m]
=f(R)
7kE[|R′|] ≥ 1− 1/e
7f(R).
Non-monotone Submodular
Before starting the analysis of Algorithm 2 for non-monotone functions, we show an
interesting property of Algorithm 1. Consistently with the notation of Section 4.2.2,
80
we use R to refer to some optimal solution. Recall that we partition the input stream
into (almost) equal-sized segments Si : 1 ≤ i ≤ k, and pick one item from each.
Then Ti denotes the set of items we have picked at the completion of segment i. We
show that f(Tk) ≥ 12ef(R ∪ Ti) for some integer i, even when f is not monotone.
Roughly speaking, the proof mainly follows from the submodularity property and
Lemma 4.2.1.
Lemma 4.2.6. If we run the monotone algorithm on a (possibly non-monotone)
submodular function f , we obtain f(Tk) ≥ 12e2f(R ∪ Ti) for some i.
Proof. Consider the stage i + 1 in which we want to pick an item from Si+1.
Lemma 4.2.1 implies
f(R ∪ Ti) ≤ f(Ti) +∑
a∈R\Ti
f(Ti ∪ a)− f(Ti).
At least one of the two right-hand side terms has to be larger than f(R ∪ Ti)/2. If
this happens to be the first term for any i, we are done: f(Tk) ≥ f(Ti) ≥ 12f(R ∪ Ti)
since f(Tk) ≥ f(Ti) by the definition of the algorithm: the first if statement makes
sure f(Ti) values are non-decreasing. Otherwise assume that the lower bound occurs
for the second terms for all values of i.
Consider the events that among the elements in R \ Ti exactly one, say a, falls
in Si+1. Call this event Ea. Conditioned on Ea, ∆i+1 := f(Ti+1) − f(Ti) is at least
f(Ti∪a)−f(Ti) with probability 1/e: i.e., if the algorithm picks the best secretary
in this interval. Each event Ea occurs with probability at least 1k· 1e. Since these
events are disjoint, we have
E[∆i+1] ≥∑
a∈R\Ti
Pr[Ea] ·1
e[f(Ti+1)− f(Ti)] ≥
1
e2k
∑a∈R\Ti
f(Ti ∪ a)− f(Ti)
≥ 1
2e2kf(R ∪ Ti),
81
and by summing over all values of i, we obtain
E[f(Tk)] =∑i
E[∆i] ≥∑i
1
2e2kf(R ∪ Ti) ≥
1
2e2minif(R ∪ Ti).
Unlike the case of monotone functions, we cannot say that f(R∪Ti) ≥ f(R), and
conclude that our algorithm is constant-competitive. Instead, we need to use other
techniques to cover the cases that f(R ∪ Ti) < f(R). The following lemma presents
an upper bound on the value of the optimum.
Lemma 4.2.7. For any pair of disjoint sets Z and Z ′, and a submodular function f ,
we have f(R) ≤ f(R ∪ Z) + f(R ∪ Z ′).
Proof. The statement follows from the submodularity property, observing that (R ∪
Z) ∩ (R ∪ Z ′) = R, and f([R ∪ Z] ∪ [R ∪ Z ′]) ≥ 0.
We are now at a position to prove the performance guarantee of our main algo-
rithm.
Theorem 4.2.8. Algorithm 2 has competitive ratio 8e2.
Proof. Let the outputs of the two algorithms be sets Z and Z ′, respectively. The
expected value of the solution is thus [f(Z) + f(Z ′)]/2.
We know that E[f(Z)] ≥ c′f(R ∪ X1) for some constant c′, and X1 ⊆ U1. The
only difference in the proof is that each element of R \ Z appears in the set Si
with probability 1/2k instead of 1/k. But we can still prove the above lemma for
c′ := 1/4e2. Same holds for Z ′: E[f(Z ′)] ≥ 14ef(R ∪X2) for some X2 ⊆ U2.
Since U1 and U2 are disjoint, so are X1 and X2. Hence, the expected value of our
solution is at least 14e2
[f(R ∪X1) + f(R ∪X2)]/2, which via Lemma 4.2.7 is at least
18e2f(R).
4.3 The Submodular Matroid Secretary Problem
In this section, we prove Theorem 4.1.2. We first design an O(log2 r)-competitive
algorithm for maximizing a monotone submodular function, when there are matroid
82
constraints for the set of selected items. Here we are allowed to choose a subset of
items only if it is an independent set in the given matroid.
The matroid (U , I) is given by an oracle access to I. Let n denote the number
of items, i.e., n := |U|, and r denotes the rank of the matroid. Let S ∈ I denote an
optimal solution that maximizes the function f . We focus our analysis on a refined
set S∗ ⊆ S that has certain nice properties: 1) f(S∗) ≥ (1 − 1/e)f(S), and 2)
f(T ) ≥ f(S∗)/ log r for any T ⊆ S∗ such that |T | = b|S∗|/2c. We cannot necessarily
find S∗, but we prove that such a set exists.
Start by letting S∗ = S. As long as there is a set T violating the second property
above, remove T from S∗, and continue. The second property clearly holds at the
termination of the procedure. In order to prove the first property, consider one
iteration. By submodularity (subadditivity to be more precise) we have f(S∗ \ T ) ≥
f(S∗)− f(T ) ≥ (1− 1/ log r)f(S∗). Since each iteration halves the set S∗, there are
at most log r iterations. Therefore, f(S∗) ≥ (1− 1/ log r)log r · f(S) ≥ (1− 1/e)f(S).
We analyze the algorithm assuming the parameter |S∗| is given, and achieve a
competitive ratio O(log r). If |S∗| is unknown, though, we can guess its value (from a
pool of log r different choices) and continue with Lemma 4.3.1. This gives an O(log2 r)
competitive ratio.
Lemma 4.3.1. Given |S∗|, Algorithm 3 picks an independent subset of items with
size |S∗|/2 whose expected value is at least f(S∗)/4e log r.
Proof. Let k := |S∗|. We divide the input stream of n items into k segments of
(almost) equal size. We only pick k/2 items, one from each of the first k/2 segments.
Similarly to Algorithm 1 for the submodular secretary problem, when we work
on each segment, we try to pick an item that maximizes the marginal value of the
function given the previous selection is fixed (see the for loop in Algorithm 1). We
show that the expected gain in each of the first k/2 segments is at least a constant
fraction of f(S∗)/k log r.
Suppose we are working on segment i ≤ k/2, and let Z be the set of items already
picked; so |Z| ≤ i−1. Furthermore, assume f(Z) ≤ f(S∗)/2 log r since otherwise, the
83
Algorithm 3 Monotone Submodular Secretary Algorithm with Matroid constraint
Input: A monotone submodular function f : 2U 7→ R, a matroid (U , I), and arandomly permuted stream of secretaries, denoted by (a1, a2, . . . , an).Output: A subset of secretaries that are independent according to I.
Let U1 := a1, a2, . . . , abn/2cPick the parameter k := |S∗| uniformly at random from the pool 20, 21, 2log rif k = O(log r) then
Select the best item of the U1 and output the singletonelse run Algorithm 1 on U1 and respect the matroid
Let T0 ← ∅Let l← bn/kcfor i← 1 to k do phase i
Let ui ← (i− 1)l + l/eLet αi ← max
(i−1)l≤j<uiTi−1∪aj∈I
f(Ti−1 ∪ aj)
if αi < f(Ti−1) thenαi ← f(Ti−1)
end ifPick an index pi : ui ≤ pi < il such that f(Ti−1∪api) ≥ αi and Ti−1∪api ∈Iif such an index pi exists then
Let Ti ← Ti−1 ∪ apielse
Let Ti ← Ti−1end if
end forOutput Tk as the solution
end if
84
lemma is already proved. By matroid properties we know there is a set T ⊆ S∗ \Z of
size bk/2b such that T ∪Z ∈ I. The second property of S∗ gives f(T ) ≥ f(S∗)/ log r.
From Lemma 4.2.1 and monotonicity of f , we obtain
∑s∈T
[f(Z ∪ s)− f(Z)] ≥ f(T ∪ Z)− f(Z) ≥ f(T )− f(Z) ≥ f(S∗)/2 log r.
Note that each item in T appears in this segment with probability 2/k because we
divided the input stream into k/2 equal segments. Since in each segment we pick the
item giving the maximum marginal value with probability 1/e, the expected gain in
this segment is at least
∑s∈T
1
e· 2
k· [f(Z ∪ s)− f(Z)] ≥ f(S∗)/ek log r.
We have this for each of the first k/2 segments, so the expected value of our solution
is at least f(S∗)/2e log r.
Finally, it is straightforward (and hence the details are omitted) to combine the
algorithm in this section with Algorithm 2 for the non-monotone submodular sec-
retary problem, to obtain an O(log2 r)-competitive algorithm for the non-monotone
submodular secretary problem subject to a matroid constraint.
Here we show the same algorithm works when there are l ≥ 1 matroid constraints
and achieves a competitive ratio of O(l log2 r). We just need to respect all matroid
constraints in Algorithm 3. This finishes the proof of Theorem 4.1.2.
Lemma 4.3.2. Given |S∗|, Algorithm 3 picks an independent subset of items (i.e.,
independent with respect to all matroids) with expected value at least f(S∗)/4el log r.
Proof. The proof is similar to the proof of Lemma 4.3.1. We show that the expected
gain in each of the first k/2l segments is at least a constant fraction of f(S∗)/k log r.
Suppose we are working on segment i ≤ k/2l, and let Z be the set of items
already picked; so |Z| ≤ i − 1. Furthermore, assume f(Z) ≤ f(S∗)/2 log r since
otherwise, the lemma is already proved. We claim that there is a set T ⊆ S∗ \ Z of
85
size k − l× bk/2lc ≥ k/2 such that T ∪ Z is an independent set in all matroids. The
proof is as follows. We know that there exists a set T1 ⊆ S∗ whose union with Z is an
independent set of the first matroid, and the size of T1 is at least |S∗|− |Z|. This can
be proved by the exchange property of matroids, i.e., adding Z to the independent
set S∗ does not remove more than |Z| items from S∗. Since T1 is independent with
respect to the second matroid (as it is a subset of S∗), we can prove that there exists
a set T2 ⊆ T1 of size at least |T1| − |Z| such that Z ∪ T2 is an independent set in the
second matroid. If we continue this process for all matroid constraints, we can prove
that there is a set Tl which is an independent set in all matroids, and has size at least
|S∗| − l|Z| ≥ k− l×bk/2lc ≥ k/2 such that Z ∪ Tl is independent with respect to all
the given matroids. The rest of the proof is similar to the proof of Lemma 4.3.1—we
just need to use the set Tl instead of the set T in the proof.
Since we are gaining a constant times f(S∗)/k log r in each of the first k/2l
segments, the expected value of the final solution is at least a constant times
f(S∗)/l log r.
4.4 Knapsack Constraints
In this section, we prove Theorem 4.1.3. We first outline how to reduce an instance
with multiple knapsacks to an instance with only one knapsack, and then we show
how to solve the single knapsack instance.
Without loss of generality, we can assume that all knapsack capacities are equal to
one. Let I be the given instance with the value function f , and item weights wij for
1 ≤ i ≤ l and 1 ≤ j ≤ n. Define a new instance I ′ with one knapsack of capacity one
in which the weight of the item j is w′j := maxiwij. We first prove that this reduction
loses no more than a factor 4l in the total value. Take note that both the scaling and
the weight transformation can be carried in an online manner as the items arrive.
Hence, the results of this section hold for the online as well as the offline setting.
Lemma 4.4.1. With instance I ′ defined above, we have 14l
OPT(I) ≤ OPT(I ′) ≤
OPT(I).
86
Proof. The latter inequality is very simple: Take the optimal solution to I ′. This is
also feasible in I since all the item weights in I are bounded by the weight in I ′.
We next prove the other inequality. Let T be the optimal solution of I. An item
j is called fat if w′j ≥ 1/2. Notice that there can be at most 2l fat items in T since∑j∈T w
′j ≤
∑j∈T∑
iwij ≤ l. If there is any fat item with value at least OPT(I)/4l,
the statement of the lemma follows immediately, so we assume this is not the case.
The total value of the fat items, say F , is at most OPT(I)/2. Submodularity and
non-negativity of f gives f(T \F ) ≥ f(T )−f(F ) ≥ OPT(I)/2. Sort the non-fat items
in decreasing order of their value density (i.e., ratio of value to weight), and let T ′ be
a maximal prefix of this ordering that is feasible with respect to I ′. If T ′ = T \ F ,
we are done; otherwise, T ′ has weight at least 1/2. Let x be the total weight of
items in T ′ and let y indicate the total weight of items in T \ (F ∪ T ′). Let αx and
αy denote the densities of the two corresponding subsets of the items, respectively.
Clearly x+ y ≤ l and αx ≥ αy. Thus, f(T \ F ) = αx · x+ αy · y ≤ αx(x+ y) ≤ αx · l.
Now f(T ′) ≥ αx · 12 ≥12lf(T \ F ) ≥ 1
4lf(T ) finishes the proof.
Here we show how to achieve a constant competitive ratio when there is only one
knapsack constraint. Let wj denote the weight of item j : 1 ≤ j ≤ n, and assume
without loss of generality that the capacity of the knapsack is 1. Moreover, let f
be the value function which is a non-monotone submodular function. Let T be the
optimal solution, and define OPT := f(T ). The value of the parameter λ ≥ 1 will be
fixed below. Define T1 and T2 as the subsets of T that appear in the first and second
half of the input stream, respectively. We first show the this solution is broken into
two balanced portions.
Lemma 4.4.2. If the value of each item is at most OPT /λ, for sufficiently large λ,
the random variable |f(T1)−f(T2)| is bounded by OPT /2 with a constant probability.
Proof. Each item of T goes to either T1 or T2 with probability 1/2. Let the random
variable X1j denote the increase of the value of f(T1) due to the possible addition
of item j. Similarly X2j is defined for the same effect on f(T2). The two variables
X1j and X2
j have the same probability distribution, and because of submodularity
87
and the fact that the value of item j is at most OPT/λ, the contribution of item j
in f(T1) − f(T2) can be seen as a random variable that always take values in range
[−OPT /λ,OPT /λ] with mean zero. (In fact, we also use the fact that in an optimal
solution, the marginal value of any item is non-negative. Submodularity guarantees
that this holds with respect to any of the subsets of T as well.) Azuma’s inequality
ensures that with constant probability the value of |f(T1)− f(T2)| is not more than
maxf(T1), f(T2)/2 for sufficiently large λ. Since both f(T1) and f(T2) are at most
OPT, we can say that they are both at least OPT /4, with constant probability.
The algorithm is as follows. Without loss of generality assume that all items are
feasible, i.e., any one item fits into the knapsack. We flip a coin, and if it turns up
“heads,” we simply try to pick the one item with the maximum value. We do the
following if the coin turns up “tails.” We do not pick any item from the first half of
the stream. Instead, we compute the maximum value set in the first half with respect
to the knapsack constraint; Lee et al. give a constant factor approximation for this
task. From the above argument, we know that f(T1) is at least OPT/4 since all the
items have limited value in this case (i.e., at most OPT /λ). Therefore, we obtain a
constant factor estimation of OPT by looking at the first half of the stream: i.e., if the
estimate is ˆOPT, we get OPT /c ≤ ˆOPT ≤ OPT. After obtaining this estimate, we
go over the second half of the input, and pick an item j if and only if it is feasible to
pick this item, and moreover, the ratio of its marginal value to wj is at least ˆOPT/6.
Lemma 4.4.3. The above algorithm is a constant competitive algorithm for the non-
monotone submodular secretary problem with one knapsack constraint.
Proof. We give the proof for the monotone case. Extending it for the non-monotone
requires the same idea as was used in the proof of Theorem 2. First suppose there is
an item with value at least OPT /λ. With probability 1/2, we try to pick the best
item, and we succeed with probability 1/e. Thus, we get an O(1) competitive ratio
in this case.
In the other case, all the items have small contributions to the solution, i.e., less
than OPT /λ. In this case, with constant probability, both f(T1) and f(T2) are at
88
least OPT /4. Hence, ˆOPT is a constant estimate for OPT. Let T ′ be the set of items
picked by the algorithm in this case. If the sum of the weights of the items in T ′ is at
least 1/2, we are done, because all these items have (marginal) value density at least
ˆOPT/6, so f(T ′) ≥ (1/2) · ( ˆOPT/6) = ˆOPT/12 ≥ OPT /48.
Otherwise, the total weight of T ′ is less than 1/2. Therefore, there are items in
T2 that are not picked. There might be two reasons for this. There was not enough
room in the knapsack, which means that the weight of the items in T2 is more than
1/2. However, there cannot be more than one such item in T2, and the value of this
item is not more than OPT/λ. Let z be this single big item, for future reference.
Therefore, f(T ′) ≥ f(T2)−OPT /λ in this case.
The other case is when the ratios of some items from T2 are less than ˆOPT/6, and
thus we do not pick them. Since they are all in T2, their total weight is at most 1.
Because of submodularity, the total loss due to these missed items is at most ˆOPT/6.
Submodularity and non-negativity of f then gives f(T ′) ≥ f(T2)−f(z)− ˆOPT/6 ≥ˆOPT−OPTλ− ˆOPT/6 = O(OPT).
4.5 The Subadditive Secretary Problem
In this section, we prove Theorem 4.1.4 by presenting first a hardness result for
approximation subadditive functions in general. The result applies in particular to
our online setting. Surprisingly, the monotone subadditive function that we use here
is almost submodular ; see Proposition 4.5.4 below. Hence, our constant competitive
ratio for submodular functions is nearly the most general we can achieve.
Definition 4.5.1 (Subadditive function maximization). Given a nonnegative subad-
ditive function f on a ground set U , and a positive integer k ≤ |U |, the goal is to
find a subset S of U of size at most k so as to maximize f(S). The function f is
accessible through a value oracle.
89
4.5.1 Hardness Result
In the following discussion, we assume that there is an upper bound of m on the size
of sets given to the oracle. We believe this restriction can be lifted. If the function
f is not required to be monotone, this is quite easy to have: simply let the value of
the function f be zero for queries of size larger than m. Furthermore, depending on
how we define the online setting, this may not be an additional restriction here. For
example, we may not be able to query the oracle with secretaries that have already
been rejected.
The main result of the section is the following theorem. It shows the subadditive
function maximization is difficult to approximate, even in the offline setting.
Theorem 4.5.2. There is no polynomial time algorithm to approximate an instance
of subadditive function maximization within O(√n) of the optimum. Furthermore,
no algorithm with exponential time 2t can achieve an approximation ratio better than
O(√n/t).
First, we are going to define our hard function. Afterwards, we continue with
proving certain properties of the function which finally lead to the proof of Theo-
rem 4.5.2.
Let n denote the size of the universe, i.e., n := |U |. Pick a random subset S∗ ⊆ U
by sampling each element of U with probability k/n. Thus, the expected size of S∗
is k.
Define the function g : U 7→ N as g(S) := |S ∩ S∗| for any S ⊆ U . One can easily
verify that g is submodular. We have a positive r whose value will be fixed below.
Define the final function f : U 7→ N as
f(S) :=
1 if g(S) = 0
dg(S)/re otherwise.
It is not difficult to verify the subadditivity of f ; it is also clearly monotone.
In order to prove the core of the hardness result in Lemma 4.5.3, we now let
90
r := λ · mkn
, where λ ≥ 1 +√
3tnmk
and t = Ω(log n) will be determined later.
Lemma 4.5.3. An algorithm making at most 2t queries to the value oracle cannot
solve the subadditive maximization problem to within k/r approximation factor.
Proof. Note that for any X ⊆ U , f(X) lies between 0 and dk/re. In fact, the optimal
solution is the set S∗ whose value is at least k/r. We prove that with high probability
the answer to all the queries of the algorithm is one. This implies that the algorithm
cannot achieve an approximation ratio better than k/r.
Assume that Xi is the i-th query of the algorithm for 1 ≤ i ≤ 2t. Notice that
Xi can be a function of our answers to the previous queries. Define Ei as the event
f(Xi) = 1. This is equivalent to g(Xi) ≤ r. We show that with high probability all
events Ei occur.
For any 1 ≤ i ≤ 2t, we have
Pr
[Ei|
i−1⋂j=1
Ej
]=
Pr[⋂ij=1 Ej]
Pr[⋂i−1j=1 Ej]
≥ Pr
[i⋂
j=1
Ej
]≥ 1−
i∑j=1
Ej.
Thus, we have Pr[∩2ti=1Ei] ≥ 1 − 2t∑2t
i=1 Pr[Ei] from union bound. Next we bound
Pr[Ei]. Consider a subset X ⊆ U such that |X| ≤ m. Since the elements of S∗ are
picked randomly with probability k/n, the expected value of X∩S∗ is at most mk/n.
Standard application of Chernoff bounds gives
Pr[f(X) 6= 1] = Pr[g(X) > r] = Pr
[|X ∩ S∗| > λ · mk
n
]≤ exp
− (λ− 1)2
mk
n
≤ exp−3t ≤ 2−2t
n,
where the last inequality follows from t ≥ log n. Therefore, the probability of all Eievents occurring simultaneously is at least 1− 1/n.
Now we can prove the main theorem of the section.
Proof of Theorem 4.5.2. We just need to set k = m =√n. Then, λ =
√3t, and the
inapproximability ratio is Ω(√
nt). Restricting to polynomial algorithms, we obtain
91
t := O(log1+ε n), and considering exponential algorithms with running time O(2t′),
we have t = O(t′), giving the desired results.
In case the query size is not bounded, we can define f(X) := 0 for large sets X,
and pull through the same result; however, the function f is no longer monotone in
this case.
We now show that the function f is almost submodular. Recall that a function g
is submodular if and only if g(A) + g(B) ≥ g(A ∪B) + g(A ∩B).
Proposition 4.5.4. For the hard function f defined above, f(A) + f(B) ≥ f(A ∪
B) + f(A ∩ B) − 2 always holds; moreover, f(X) is always positive and attains a
maximum value of Θ(√n) for the parameters fixed in the proof of Theorem 4.5.2.
Proof. The function h(X) := g(X)/r is clearly submodular, and we have h(X) ≤
f(X) ≤ h(X) + 1. We obtain f(A) + f(B) ≥ h(A) + h(B) ≥ h(A∪B) + h(A∩B) ≥
f(A ∪B) + f(A ∩B)− 2.
4.5.2 Algorithm
An algorithm that only picks the best item clearly gives a k competitive ratio. We now
show how to achieve an O(n/k) competitive ratio, and thus by combining the two,
we obtain an O(√n)-competitive algorithm for the monotone subadditive secretary
problem. This result complements our negative result nicely.
Partition the input stream S into ` := n/k (almost) equal-sized segments, each
of size at most k. Randomly pick all the elements in one of these segments. Let the
segments be denoted by S1, S2, . . . , S`. Subadditivity of f implies f(S) ≤∑
i f(Si).
Hence, the expected value of our solution is∑
i1`f(Si) ≥ 1
`f(S) ≥ 1
`OPT, where the
two inequalities follow from subadditivity and monotonicity, respectively.
4.6 Further Results
In this chapter, we consider the (non-monotone) submodular secretary problem for
which we give a constant-competitive algorithm. The result can be generalized when
92
we have a matroid constraint on the set that we pick; in this case we obtain an
O(log2 r)-competitive algorithm where r is the rank of the matroid. However, we
show that it is very hard to compete with the optimum if we consider subadditive
functions instead of submodular functions. This hardness holds even for “almost
submodular” functions; see Proposition 4.5.4.
One may consider special non-submodular functions which enjoy certain structural
results in order to find better guarantees. For example, let f(T ) be the minimum
individual value in T which models a bottle-neck situation in the secretary problem,
i.e., selecting a group of k secretaries to work together, and the speed (efficiency) of
the group is limited to that of the slowest person in the group (note that unlike the
submodular case here the condition for employing exactly k secretaries is enforced.)
In this case, we present a simple O(k) competitive ratio for the problem as follows.
Interview the first 1/k fraction of the secretaries without employing anyone. Let α be
the highest efficiency among those interviewed. Employ the first k secretaries whose
efficiency surpasses α.
Theorem 4.6.1. Following the prescribed approach, we employ the k best secretaries
with probability at least 1/e2k.
Indeed we believe that this O(k) competitive ratio for this case should be almost
tight. One can verify that provided individual secretary efficiencies are far from each
other, say each two consecutive values are farther than a multiplicative factor n, the
problem of maximizing the expected value of the minimum efficiency is no easier than
being required to employ all the k best secretaries. The following theorem provides
evidence that the latter problem is hard to approximate.
Theorem 4.6.2. Any algorithm with a single threshold—i.e., interviewing applicants
until some point (observation phase), and then employing any one who is better than
all those in the observation phase—misses one of the k best secretaries with probability
1−O(log k/k).
Another important aggregation function f is that of maximizing the performance
of the secretaries we employ: think of picking k candidate secretaries and finally
93
hiring the best. We consider this function in Subsection 4.6.1 for which we present a
near-optimal solution. In fact, the problem has been already studied, and an optimal
strategy appears in [40]. However, we propose a simpler solution which features
certain “robustness” properties (and thus is of its own interest): in particular, suppose
we are given a vector (γ1, γ2, . . . , γk) such that γi ≥ γi+1 for 1 ≤ i < k. Sort the
elements in a set R of size k in a non-increasing order, say a1, a2, . . . , ak. The goal
is to maximize the efficiency∑
i γiai. The algorithm that we propose maximizes this
more general objective obliviously; i.e., the algorithm runs irrespective of the vector
γ, however, it can be shown the resulting solution approximates the objective for all
vectors γ at the same time. We refer to Subsection 4.6.1 for more details. Following
we provide proofs of theorems 4.6.1 and, 4.6.2.
Proof of Theorem 4.6.1. Let R = a1, a2, . . . , a|R| ⊆ S denote the set of k best
secretaries. Let S∗ denote the first 1/k fraction of the stream of secretaries. Let E1
denote the event when S∗ ∩ R = ∅, that is, we do not lose the chance of employing
the best secretaries (R) by being a mere observer in S∗. Let E2 denote the event that
we finally pick the set R. Let us first bound Pr[E1]. In order to do so, define E1j for
j : 1 ≤ j ≤ |R| as the event that aj 6∈ Se. We know that Pr[E11 ] ≥ 1/k. In general,
we have for j > 1
Pr
[E1j
∣∣∣∣∣⋂i<j
E1i
]≥n− n
k− j + 1
n− j + 1
≥n− n
k− k
n− k
= 1− n/k
n− k
≥ 1− 2
kassuming k ≤ n
2. (4.4)
Notice that the final assumption is justified because we can solve the problem of
finding the k′ = n−k ≤ n/2 smallest numbers in case k > n/2. Using Equation (4.4)
94
we obtain
Pr[E1] = Pr[E11 ] Pr[E12 |E11 ] · · ·Pr[E1|R|| ∪j<|R| E1j ]
≥(
1− 2
k
)k≥ e−2. (4.5)
The event E2 happens when E1 happens and the (k+ 1)th largest element appears in
S∗. Thus, we have Pr[E2] = Pr[E1] Pr[E2|E1] ≥ e−2 · 1/k = 1e2k
.
Proof of Theorem 4.6.2. We assume that we cannot find the actual efficiency of a
secretary, but we only have an oracle that given two secretaries already interviewed,
reports the better of the two. This model is justified if the range of efficiency values
is large, and a suitable perturbation is introduced into the values.
Suppose the first secretary is hired after interviewing a β fraction of the secretaries.
If β > log k/k then the probability that we miss at least one of the k best secretaries
is at least 1− (1− β)k = 1− 1/k. If on the other hand, β is small, say β ≤ log k/k,
there is little chance that the right threshold can be picked. Notice that in the oracle
model, the threshold has to be the efficiency of one prior secretary. Thus for the right
threshold to be selected, we need to have the (k + 1)th best secretary in the first β
fraction—the probability of this even is no more than β. Therefore, the probability
of success cannot be more than log k/k.
4.6.1 The Secretary Problem with the “Maximum” Function
We now turn to consider a different efficiency aggregation function, namely the max-
imum of the efficiency of the individual secretaries. Alternately, one can think of this
function as a secretary function with k choices, that is, we select k secretaries and we
are satisfied as long as one of them is the best secretary interviewed. We propose an
algorithm that accomplishes this task with probability 1−O(ln kk
)for k > 1.
As we did before, we assume that n is a multiple of k, and we partition the
input stream into k equally-sized segments, named S1, S2, . . . , Sk. Let f(s) denote
95
the efficiency of the secretary s ∈ S. For each set i : 1 ≤ i < k, we compute
αi := maxs∈
⋃j≤i Si
f(s),
which is the efficiency of the best secretary in the first i segments. Clearly, αi can
be computed in an online manner after interviewing the first i segments. For each
i : 1 ≤ i < k, we try to employ the first secretary in⋃j>i Sj whose efficiency surpasses
αi. Let this choice, if at all present, be denoted si. The output of the algorithm
consists of all such secretaries sii. Notice that such an element may not exist for a
particular i, or we may have si = si′ for i 6= i′. We employ at most k − 1 secretaries.
The following theorem bounds the failure probability of the algorithm.
Theorem 4.6.3. The probability of not employing the best secretary is O(ln kk
).
Proof. Let (a1, a2, . . . , an) denote the stream of interviewed secretaries. Let am be
the best secretary, and suppose am ∈ Si, namely (i − 1)l < m ≤ il, where l := n/k.
Our algorithm is successful if the second best secretary of the set a1, a2, . . . , am−1
does not belong to Si. The probability of this event is
(i− 1)l
m≥ (i− 1)l
il=i− 1
i. (4.6)
The probability of am ∈ Si is 1/k and conditioned on this event, the probability of
failure is at most 1/i. Hence, the total failure probability is no more than∑k
i=11k1i
=
O(ln kk
)as claimed.
This problem has been previously studied by Gilbert and Mosteller [40]. Our
algorithm above is simpler and yet “robust” in the following sense. The primary
goal is to select the best secretary, but we also guarantee that many of the “good”
secretaries are also selected. In particular, we show that the better the rank of a
secretary is in our evaluation, the higher is the guarantee we have for employing her.
Theorem 4.6.4. The probability of not hiring a secretary of rank y is O(√
yk
).
96
Proof. Let (a1, a2, . . . , an) denote the stream of interviewed secretaries. Let am be the
secretary of rank y, and suppose am ∈ Si, namely (i− 1)l < m ≤ il, where l := n/k.
Below we define three bad events whose probabilities we bound, and we show that am
is hired provided none of these events occur. In particular, we give an upper bound
of O(√y/k for each event. The claim then follows from the union bound.
Let z :=√
ky−1 − 1. The event E1 occurs if i ≤ z. This event happens with
probability z/k which is less than√
1k(y−1) ≤
√yk.
We say the event E2 happens if am is not the best secretary among those in sets
Si, Si−1, . . . , Si−z. This happens when there is at least one of the y − 1 secretaries
better than am in these sets. Let W be a random variable denoting the number of
these y − 1 secretaries in any of the mentioned sets. Since any secretary is in one
of these sets with probability (z + 1)/k (notice that z + 1 is the number of these
sets), we can say that the expected value of W is (y− 1)(z+ 1)/k. Using the Markov
Inequality, the probability that W is at least 1 is at most its expected value which is
(y−1)(z+1)/k. Thus, using the definition of z, we get an upper bound of O(√
y−1k
)for E2.
Finally, we define E3 as the event when the best secretary among
a(i−z−1)l+1, a(i−z−1)l+2, . . . , aj−1 (secretaries appearing before am in the above-
mentioned sets) is in set Si. This happens with probability at most 1/(z+1), because
there are z + 1 sets that the best secretary is equally likely to be in each. Thus, we
get Pr[E3] = O(√
yk
)by definition of z.
If non of the events E1, E2 and E3 happen, we claim am is employed. Because if
the maximum item of items a(i−z−1)l+1, a(i−z−1)l+2, . . . , aj−1 is in the set Si′ , and
i− z ≤ i′ < i, then we hire am for the set Si′ ; refer to the algorithm when we consider
the threshold αi′ .
The aforementioned algorithm of [40] misses a good secretary of rank y with prob-
ability roughly 1/y. On the other hand, one can show that the algorithm of Klein-
berg [59] (for maximizing the sum of the efficiencies of the secretaries) picks secretaries
of high rank with probability about 1−Θ(1/√k). However, the latter algorithm guar-
antees the selection of the best secretary with a probability no more than O(1/√k).
97
Therefore, our algorithm has the nice features of both these algorithms: the best
secretary is hired with a very good probability, while other good secretaries also have
a good chance of being employed.
98
Bibliography
[1] A. A. Ageev and M. I. Sviridenko. An 0.828-approximation algorithm for the
uncapacitated facility location problem. Discrete Appl. Math., 93(2-3):149–156,
1999.
[2] S. Agrawal, Z. Wang, and Y. Ye. A dynamic near-optimal algorithm for online
linear programming. Working paper posted at http://www.stanford.edu/ yyye/.
[3] Miklos Ajtai, Nimrod Megiddo, and Orli Waarts. Improved algorithms and
analysis for secretary problems and generalizations. SIAM J. Discrete Math.,
14(1):1–27, 2001.
[4] David Applegate and Edith Cohen. Making routing robust to changing traffic
demands: algorithms and evaluation. IEEE/ACM Trans. Netw., 16(6):1193–
1206, 2006.
[5] Arash Asadpour, Hamid Nazerzadeh, and Amin Saberi. Stochastic submodular
maximization. In WINE, pages 477–489, 2008.
[6] Yossi Azar, Edith Cohen, Amos Fiat, Haim Kaplan, and Harald Racke. Optimal
oblivious routing in polynomial time. J. Comput. Syst. Sci., 69(3):383–394, 2004.
[7] Moshe Babaioff, Nicole Immorlica, David Kempe, and Robert Kleinberg. A
knapsack secretary problem with applications. In APPROX, pages 16–28, 2007.
[8] Moshe Babaioff, Nicole Immorlica, David Kempe, and Robert Kleinberg. Online
auctions and generalized secretary problems. SIGecom Exch., 7(2):1–11, 2008.
99
[9] Moshe Babaioff, Nicole Immorlica, and Robert Kleinberg. Matroids, secretary
problems, and online mechanisms. In SODA, pages 434–443, 2007.
[10] Bahman Bahmani and Michael Kapralov. Improved bounds for online stochastic
matching. In ESA, pages 170–181, 2010.
[11] MohammadHossein Bateni, Mohammad Taghi Hajiaghayi, and Morteza Zadi-
moghaddam. Submodular secretary problem and extensions. ACM Transactions
on Algorithms, 9(4):32, 2013.
[12] MohammadHossein Bateni, MohammadTaghi Hajiaghayi, and Morteza Zadi-
moghaddam. The submodular secretary problem. Technical Report TD-7UEP26,
AT&T Labs–Research, July 2009.
[13] Aharon Ben-Tal and Arkadi Nemirovski. Robust optimization - methodology
and applications. Math. Program., 92(3):453–480, 2002.
[14] Dimitris Bertsimas, Dessislava Pachamanova, and Melvyn Sim. Robust linear
optimization under general norms. Oper. Res. Lett., 32(6):510–516, 2004.
[15] Anand Bhalgat, Jon Feldman, and Vahab S. Mirrokni. Online ad allocation with
smooth delivery. In ACM Conference on Knowledge Discovery (KDD), 2012.
[16] V. Bilo, M. Flammini, and L. Moscardelli. Pareto approximations for the bi-
criteria scheduling problem. Journal of Parallel and Distributed Computing,
66(3):393–402, 2006.
[17] Benjamin Birnbaum and Claire Mathieu. On-line bipartite matching made sim-
ple. SIGACT News, 39(1):80–87, 2008.
[18] N. Buchbinder, K. Jain, and J.S. Naor. Online Primal-Dual Algorithms for
Maximizing Ad-Auctions Revenue. In Proc. ESA, page 253. Springer, 2007.
[19] N. Buchbinder and J.S. Naor. Fair online load balancing. In Proceedings of the
eighteenth annual ACM symposium on Parallelism in algorithms and architec-
tures, pages 291–298. ACM, 2006.
100
[20] Gruia Calinescu, Chandra Chekuri, Martin Pal, and Jan Vondrak. Maximizing
a submodular set function subject to a matroid constraint (extended abstract).
In IPCO, pages 182–196, 2007.
[21] Gerard Cornuejols, Marshall Fisher, and George L. Nemhauser. On the unca-
pacitated location problem. In Studies in integer programming (Proc. Workshop,
Bonn. 1975), pages 163–177. Ann. of Discrete Math., Vol. 1. North-Holland, Am-
sterdam, 1977.
[22] Gerard Cornuejols, Marshall L. Fisher, and George L. Nemhauser. Location of
bank accounts to optimize float: an analytic study of exact and approximate
algorithms. Manage. Sci., 23(8):789–810, 1976/77.
[23] Nikhil Devanur and Thomas Hayes. The adwords problem: Online keyword
matching with budgeted bidders under random permutations. In ACM EC,
2009.
[24] Nikhil R. Devanur, Zhiyi Huang, Nitish Korula, Vahab S. Mirrokni, and Qiqi
Yan. Whole-page optimization and submodular welfare maximization with online
bidders. In ACM Conference on Electronic Commerce, pages 305–322, 2013.
[25] Nikhil R. Devanur, Kamal Jain, Balasubramanian Sivan, and Christopher A.
Wilkens. Near optimal online algorithms and fast approximation algorithms
for resource allocation problems. In ACM Conference on Electronic Commerce,
pages 29–38, 2011.
[26] Nikhil R. Devanur, Balasubramanian Sivan, and Yossi Azar. Asymptotically
optimal algorithm for stochastic adwords. In ACM ee on Electronic Commerce,
pages 388–404, 2012.
[27] E. B. Dynkin. The optimum choice of the instant for stopping a markov process.
Sov. Math. Dokl., 4:627–629, 1963.
101
[28] Jack Edmonds. Submodular functions, matroids, and certain polyhedra. In
Combinatorial Structures and their Applications (Proc. Calgary Internat. Conf.,
Calgary, Alta., 1969), pages 69–87. Gordon and Breach, New York, 1970.
[29] U. Feige and M. X. Goemans. Approximating the value of two power proof
systems, with applications to MAX 2SAT and MAX DICUT. In ISTCS, page
182, 1995.
[30] Uriel Feige. A threshold of lnn for approximating set cover. J. ACM, 45(4):634–
652, 1998.
[31] Uriel Feige. On maximizing welfare when utility functions are subadditive. In
STOC, pages 41–50, 2006.
[32] Uriel Feige, Vahab S. Mirrokni, and Jan Vondrak. Maximizing non-monotone
submodular functions. In FOCS, pages 461–471, 2007.
[33] J. Feldman, N. Korula, V. Mirrokni, S. Muthukrishnan, and M. Pal. Online ad
assignment with free disposal. In WINE, 2009.
[34] Jon Feldman, Monika Henzinger, Nitish Korula, Vahab S. Mirrokni, and Clif-
ford Stein. Online stochastic packing applied to display ad allocation. In Mark
de Berg and Ulrich Meyer, editors, Algorithms - ESA 2010, 18th Annual Eu-
ropean Symposium, Liverpool, UK, September 6-8, 2010. Proceedings, Part I,
volume 6346 of Lecture Notes in Computer Science, pages 182–194. Springer,
2010.
[35] Jon Feldman, Nitish Korula, Vahab S. Mirrokni, S. Muthukrishnan, and Martin
Pal. Online ad assignment with free disposal. In Stefano Leonardi, editor, Inter-
net and Network Economics, 5th International Workshop, WINE 2009, Rome,
Italy, December 14-18, 2009. Proceedings, volume 5929 of Lecture Notes in Com-
puter Science, pages 374–385. Springer, 2009.
[36] Jon Feldman, Aranyak Mehta, Vahab Mirrokni, and S. Muthukrishnan. Online
stochastic matching: Beating 1 - 1/e. In FOCS, 2009.
102
[37] M. L. Fisher, G. L. Nemhauser, and L. A. Wolsey. An analysis of approximations
for maximizing submodular set functions. II. Math. Prog. Stud., (8):73–87, 1978.
Polyhedral combinatorics.
[38] M. Flammini and G. Nicosia. On multicriteria online problems. In Proceedings
of the 8th Annual European Symposium on Algorithms, pages 191–201. Springer-
Verlag, 2000.
[39] P. R. Freeman. The secretary problem and its extensions: a review. Internat.
Statist. Rev., 51(2):189–206, 1983.
[40] John P. Gilbert and Frederick Mosteller. Recognizing the maximum of a se-
quence. J. Amer. Statist. Assoc., 61:35–73, 1966.
[41] Kenneth S. Glasser, Richard Holzsager, and Austin Barron. The d choice secre-
tary problem. Comm. Statist. C—Sequential Anal., 2(3):177–199, 1983.
[42] A. Goel, A. Meyerson, and S. Plotkin. Approximate majorization and fair online
load balancing. In Proceedings of the twelfth annual ACM-SIAM symposium on
Discrete algorithms, pages 384–390. Society for Industrial and Applied Mathe-
matics, 2001.
[43] A. Goel, A. Meyerson, and S. Plotkin. Combining fairness with throughput: On-
line routing with multiple objectives. Journal of Computer and System Sciences,
63(1):62–79, 2001.
[44] Michel X. Goemans and David P. Williamson. Improved approximation algo-
rithms for maximum cut and satisfiability problems using semidefinite program-
ming. J. Assoc. Comput. Mach., 42(6):1115–1145, 1995.
[45] Anupam Gupta, Aaron Roth, Grant Schoenebeck, and Kunal Talwar. Con-
strained non-monotone submodular maximization: Offline and secretary algo-
rithms. In WINE, pages 246–257, 2010.
103
[46] B. Haeupler, V. Mirrokni, and M. Zadimoghaddam. Online stochastic weighted
matching: Improved approximation algorithms. In WINE, 2011.
[47] Mohammad T. Hajiaghayi, Robert Kleinberg, and Tuomas Sandholm. Auto-
mated online mechanism design and prophet inequalities. In AAAI, pages 58–65,
2007.
[48] Mohammad Taghi Hajiaghayi, Robert Kleinberg, and David C. Parkes. Adaptive
limited-supply online auctions. In EC, pages 71–80, 2004.
[49] Eran Halperin and Uri Zwick. Combinatorial approximation algorithms for the
maximum directed cut problem. In SODA, pages 1–7, 2001.
[50] Johan Hastad. Some optimal inapproximability results. J. ACM, 48(4):798–859,
2001.
[51] Elad Hazan, Shmuel Safra, and Oded Schwartz. On the complexity of approxi-
mating k-set packing. Computational Complexity, 15(1):20–39, 2006.
[52] Nicole Immorlica, Robert D. Kleinberg, and Mohammad Mahdian. Secretary
problems with competing employers. In WINE, pages 389–400, 2006.
[53] Satoru Iwata, Lisa Fleischer, and Satoru Fujishige. A combinatorial strongly
polynomial algorithm for minimizing submodular functions. J. ACM, 48(4):761–
777, 2001.
[54] Bala Kalyanasundaram and Kirk Pruhs. On-line network optimization problems.
In Developments from a June 1996 seminar on Online algorithms: the state of
the art, pages 268–280, London, UK, 1998. Springer-Verlag.
[55] Chinmay Karande, Aranyak Mehta, and Pushkar Tripathi. Online bipartite
matching with unknown distributions. In STOC, 2011.
[56] Richard M. Karp, Umesh V. Vazirani, and Vijay V. Vazirani. An optimal algo-
rithm for on-line bipartite matching. In STOC, pages 352–358, 1990.
104
[57] Subhash Khot, Guy Kindler, Elchanan Mossel, and Ryan O’Donnell. Optimal
inapproximability results for max-cut and other 2-variable csps? In FOCS, pages
146–154, 2004.
[58] Samir Khuller, Anna Moss, and Joseph Naor. The budgeted maximum coverage
problem. Inf. Process. Lett., 70(1):39–45, 1999.
[59] Robert Kleinberg. A multiple-choice secretary algorithm with applications to
online auctions. In SODA, pages 630–631, 2005.
[60] Nitish Korula, Vahab S. Mirrokni, and Morteza Zadimoghaddam. Bicriteria
online matching: Maximizing weight and cardinality. In to appear in Proceedings
of the Ninth Conference on Web and Internet Economics, WINE ’13, 2013.
[61] Jon Lee, Vahab Mirrokni, Viswanath Nagarajan, and Maxim Sviridenko. Max-
imizing non-monotone submodular functions under matroid and knapsack con-
straints. In STOC, pages 323–332, 2009.
[62] L. Lovasz. Submodular functions and convexity. In Mathematical programming:
the state of the art (Bonn, 1982), pages 235–257. Springer, Berlin, 1983.
[63] Mohammad Mahdian, Hamid Nazerzadeh, and Amin Saberi. Allocating online
advertisement space with unreliable estimates. In ACM Conference on Electronic
Commerce, pages 288–294, 2007.
[64] Mohammad Mahdian and Qiqi Yan. Online bipartite matching with random
arrivals: A strongly factor revealing lp approach. In STOC, 2011.
[65] Aranyak Mehta, Amin Saberi, Umesh Vazirani, and Vijay Vazirani. Adwords
and generalized online matching. J. ACM, 54(5):22, 2007.
[66] H. Menshadi, S. OveisGharan, and A. Saberi. Offline optimization for online
stochastic matching. In SODA, 2011.
105
[67] Vahab S. Mirrokni, Shayan Oveis Gharan, and Morteza Zadimoghaddam. Simul-
taneous approximations for adversarial and stochastic online budgeted allocation.
In SODA, pages 1690–1701, 2012.
[68] Rajeev Motwani, Rina Panigrahy, and Ying Xu 0002. Fractional matching via
balls-and-bins. In APPROX-RANDOM, pages 487–498, 2006.
[69] G. L. Nemhauser, L. A. Wolsey, and M. L. Fisher. An analysis of approximations
for maximizing submodular set functions. I. Math. Program., 14(3):265–294,
1978.
[70] Alessandro Panconesi and Aravind Srinivasan. Randomized distributed edge
coloring via an extension of the chernoff-hoeffding bounds. Siam Journal on
Computing, 26:350–368, 1997.
[71] Maurice Queyranne. A combinatorial algorithm for minimizing symmetric sub-
modular functions. In SODA, pages 98–101, 1995.
[72] Alexander Schrijver. A combinatorial algorithm minimizing submodular func-
tions in strongly polynomial time. J. Combin. Theory Ser. B, 80(2):346–355,
2000.
[73] Maxim Sviridenko. A note on maximizing a submodular set function subject to
a knapsack constraint. Oper. Res. Lett., 32(1):41–43, 2004.
[74] Bo Tan and R. Srikant. Online advertisement, optimization and stochastic net-
works. CoRR, abs/1009.0870, 2010.
[75] R. J. Vanderbei. The optimal choice of a subset of a population. Math. Oper.
Res., 5(4):481–486, 1980.
[76] Erik Vee, Sergei Vassilvitskii, and Jayavel Shanmugasundaram. Optimal online
assignment with forecasts. In ACM EC, 2010.
[77] Jan Vondrak. Symmetry and approximability of submodular maximization prob-
lems. In FOCS, 2009.
106
[78] C.M. Wang, X.W. Huang, and C.C. Hsu. Bi-objective optimization: An online
algorithm for job assignment. Advances in Grid and Pervasive Computing, pages
223–234, 2009.
[79] Hao Wang, Haiyong Xie, Lili Qiu, Yang Richard Yang, Yin Zhang, and Albert
Greenberg. Cope: traffic engineering in dynamic networks. In Proceedings of the
2006 conference on Applications, technologies, architectures, and protocols for
computer communications, SIGCOMM ’06, pages 99–110, New York, NY, USA,
2006. ACM.
[80] John G. Wilson. Optimal choice and assignment of the best m of n randomly
arriving items. Stochastic Process. Appl., 39(2):325–343, 1991.
107
